C3_W3_Lab_2_intro_to_gradcam__M.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Introduction To GradCAM (Part 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this lecture notebook we'll be looking at an introduction to Grad-CAM, a powerful technique for interpreting Convolutional Neural Networks. Grad-CAM stands for Gradient-weighted Class Activation Mapping.\n",
    "\n",
    "CNN's are very flexible models and their great predictive power comes at the cost of losing interpretability (something that is true for all Artificial Neural Networks). Grad-CAM attempts to solve this by giving us a graphical visualisation of parts of an image that are the most relevant for the CNN when predicting a particular class.\n",
    "\n",
    "Aside from working on some Grad-CAM concepts we'll also look at how we can use Keras to access some concrete information of our model. Let's dive into it!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import keras\n",
    "from keras import backend as K\n",
    "from util import *"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The `load_C3M3_model()` function has been taken care of and its internals are out of the scope of this notebook. But if it intrigues you, you can take a look at it in `util.py` "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Got loss weights\n",
      "Loaded DenseNet\n",
      "Added layers\n",
      "Compiled Model\n",
      "Loaded Weights\n"
     ]
    }
   ],
   "source": [
    "# Load the model we are going to be using\n",
    "path = \"./\"\n",
    "model = load_C3M3_model(path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As you may already know, we can check the architecture of our model using the `summary()` method. \n",
    "\n",
    "After running the code block below we’ll see that this model has a lot of layers. One advantage of Grad-CAM over previous attempts of interpreting CNN's (such as CAM) is that it is architecture agnostic. This means it can be used for CNN's with complex architectures such as this one:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, None, None, 3 0                                            \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_1 (ZeroPadding2D (None, None, None, 3 0           input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "conv1/conv (Conv2D)             (None, None, None, 6 9408        zero_padding2d_1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "conv1/bn (BatchNormalization)   (None, None, None, 6 256         conv1/conv[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "conv1/relu (Activation)         (None, None, None, 6 0           conv1/bn[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_2 (ZeroPadding2D (None, None, None, 6 0           conv1/relu[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "pool1 (MaxPooling2D)            (None, None, None, 6 0           zero_padding2d_2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block1_0_bn (BatchNormali (None, None, None, 6 256         pool1[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block1_0_relu (Activation (None, None, None, 6 0           conv2_block1_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block1_1_conv (Conv2D)    (None, None, None, 1 8192        conv2_block1_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block1_1_bn (BatchNormali (None, None, None, 1 512         conv2_block1_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block1_1_relu (Activation (None, None, None, 1 0           conv2_block1_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block1_2_conv (Conv2D)    (None, None, None, 3 36864       conv2_block1_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block1_concat (Concatenat (None, None, None, 9 0           pool1[0][0]                      \n",
      "                                                                 conv2_block1_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block2_0_bn (BatchNormali (None, None, None, 9 384         conv2_block1_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block2_0_relu (Activation (None, None, None, 9 0           conv2_block2_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block2_1_conv (Conv2D)    (None, None, None, 1 12288       conv2_block2_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block2_1_bn (BatchNormali (None, None, None, 1 512         conv2_block2_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block2_1_relu (Activation (None, None, None, 1 0           conv2_block2_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block2_2_conv (Conv2D)    (None, None, None, 3 36864       conv2_block2_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block2_concat (Concatenat (None, None, None, 1 0           conv2_block1_concat[0][0]        \n",
      "                                                                 conv2_block2_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block3_0_bn (BatchNormali (None, None, None, 1 512         conv2_block2_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block3_0_relu (Activation (None, None, None, 1 0           conv2_block3_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block3_1_conv (Conv2D)    (None, None, None, 1 16384       conv2_block3_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block3_1_bn (BatchNormali (None, None, None, 1 512         conv2_block3_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block3_1_relu (Activation (None, None, None, 1 0           conv2_block3_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block3_2_conv (Conv2D)    (None, None, None, 3 36864       conv2_block3_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block3_concat (Concatenat (None, None, None, 1 0           conv2_block2_concat[0][0]        \n",
      "                                                                 conv2_block3_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block4_0_bn (BatchNormali (None, None, None, 1 640         conv2_block3_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block4_0_relu (Activation (None, None, None, 1 0           conv2_block4_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block4_1_conv (Conv2D)    (None, None, None, 1 20480       conv2_block4_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block4_1_bn (BatchNormali (None, None, None, 1 512         conv2_block4_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block4_1_relu (Activation (None, None, None, 1 0           conv2_block4_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block4_2_conv (Conv2D)    (None, None, None, 3 36864       conv2_block4_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block4_concat (Concatenat (None, None, None, 1 0           conv2_block3_concat[0][0]        \n",
      "                                                                 conv2_block4_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block5_0_bn (BatchNormali (None, None, None, 1 768         conv2_block4_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block5_0_relu (Activation (None, None, None, 1 0           conv2_block5_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block5_1_conv (Conv2D)    (None, None, None, 1 24576       conv2_block5_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block5_1_bn (BatchNormali (None, None, None, 1 512         conv2_block5_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block5_1_relu (Activation (None, None, None, 1 0           conv2_block5_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block5_2_conv (Conv2D)    (None, None, None, 3 36864       conv2_block5_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block5_concat (Concatenat (None, None, None, 2 0           conv2_block4_concat[0][0]        \n",
      "                                                                 conv2_block5_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block6_0_bn (BatchNormali (None, None, None, 2 896         conv2_block5_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block6_0_relu (Activation (None, None, None, 2 0           conv2_block6_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block6_1_conv (Conv2D)    (None, None, None, 1 28672       conv2_block6_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block6_1_bn (BatchNormali (None, None, None, 1 512         conv2_block6_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block6_1_relu (Activation (None, None, None, 1 0           conv2_block6_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block6_2_conv (Conv2D)    (None, None, None, 3 36864       conv2_block6_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2_block6_concat (Concatenat (None, None, None, 2 0           conv2_block5_concat[0][0]        \n",
      "                                                                 conv2_block6_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "pool2_bn (BatchNormalization)   (None, None, None, 2 1024        conv2_block6_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "pool2_relu (Activation)         (None, None, None, 2 0           pool2_bn[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "pool2_conv (Conv2D)             (None, None, None, 1 32768       pool2_relu[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "pool2_pool (AveragePooling2D)   (None, None, None, 1 0           pool2_conv[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block1_0_bn (BatchNormali (None, None, None, 1 512         pool2_pool[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block1_0_relu (Activation (None, None, None, 1 0           conv3_block1_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block1_1_conv (Conv2D)    (None, None, None, 1 16384       conv3_block1_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block1_1_bn (BatchNormali (None, None, None, 1 512         conv3_block1_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block1_1_relu (Activation (None, None, None, 1 0           conv3_block1_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block1_2_conv (Conv2D)    (None, None, None, 3 36864       conv3_block1_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block1_concat (Concatenat (None, None, None, 1 0           pool2_pool[0][0]                 \n",
      "                                                                 conv3_block1_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block2_0_bn (BatchNormali (None, None, None, 1 640         conv3_block1_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block2_0_relu (Activation (None, None, None, 1 0           conv3_block2_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block2_1_conv (Conv2D)    (None, None, None, 1 20480       conv3_block2_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block2_1_bn (BatchNormali (None, None, None, 1 512         conv3_block2_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block2_1_relu (Activation (None, None, None, 1 0           conv3_block2_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block2_2_conv (Conv2D)    (None, None, None, 3 36864       conv3_block2_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block2_concat (Concatenat (None, None, None, 1 0           conv3_block1_concat[0][0]        \n",
      "                                                                 conv3_block2_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block3_0_bn (BatchNormali (None, None, None, 1 768         conv3_block2_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block3_0_relu (Activation (None, None, None, 1 0           conv3_block3_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block3_1_conv (Conv2D)    (None, None, None, 1 24576       conv3_block3_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block3_1_bn (BatchNormali (None, None, None, 1 512         conv3_block3_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block3_1_relu (Activation (None, None, None, 1 0           conv3_block3_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block3_2_conv (Conv2D)    (None, None, None, 3 36864       conv3_block3_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block3_concat (Concatenat (None, None, None, 2 0           conv3_block2_concat[0][0]        \n",
      "                                                                 conv3_block3_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block4_0_bn (BatchNormali (None, None, None, 2 896         conv3_block3_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block4_0_relu (Activation (None, None, None, 2 0           conv3_block4_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block4_1_conv (Conv2D)    (None, None, None, 1 28672       conv3_block4_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block4_1_bn (BatchNormali (None, None, None, 1 512         conv3_block4_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block4_1_relu (Activation (None, None, None, 1 0           conv3_block4_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block4_2_conv (Conv2D)    (None, None, None, 3 36864       conv3_block4_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block4_concat (Concatenat (None, None, None, 2 0           conv3_block3_concat[0][0]        \n",
      "                                                                 conv3_block4_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block5_0_bn (BatchNormali (None, None, None, 2 1024        conv3_block4_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block5_0_relu (Activation (None, None, None, 2 0           conv3_block5_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block5_1_conv (Conv2D)    (None, None, None, 1 32768       conv3_block5_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block5_1_bn (BatchNormali (None, None, None, 1 512         conv3_block5_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block5_1_relu (Activation (None, None, None, 1 0           conv3_block5_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block5_2_conv (Conv2D)    (None, None, None, 3 36864       conv3_block5_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block5_concat (Concatenat (None, None, None, 2 0           conv3_block4_concat[0][0]        \n",
      "                                                                 conv3_block5_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block6_0_bn (BatchNormali (None, None, None, 2 1152        conv3_block5_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block6_0_relu (Activation (None, None, None, 2 0           conv3_block6_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block6_1_conv (Conv2D)    (None, None, None, 1 36864       conv3_block6_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block6_1_bn (BatchNormali (None, None, None, 1 512         conv3_block6_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block6_1_relu (Activation (None, None, None, 1 0           conv3_block6_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block6_2_conv (Conv2D)    (None, None, None, 3 36864       conv3_block6_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block6_concat (Concatenat (None, None, None, 3 0           conv3_block5_concat[0][0]        \n",
      "                                                                 conv3_block6_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block7_0_bn (BatchNormali (None, None, None, 3 1280        conv3_block6_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block7_0_relu (Activation (None, None, None, 3 0           conv3_block7_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block7_1_conv (Conv2D)    (None, None, None, 1 40960       conv3_block7_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block7_1_bn (BatchNormali (None, None, None, 1 512         conv3_block7_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block7_1_relu (Activation (None, None, None, 1 0           conv3_block7_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block7_2_conv (Conv2D)    (None, None, None, 3 36864       conv3_block7_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block7_concat (Concatenat (None, None, None, 3 0           conv3_block6_concat[0][0]        \n",
      "                                                                 conv3_block7_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block8_0_bn (BatchNormali (None, None, None, 3 1408        conv3_block7_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block8_0_relu (Activation (None, None, None, 3 0           conv3_block8_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block8_1_conv (Conv2D)    (None, None, None, 1 45056       conv3_block8_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block8_1_bn (BatchNormali (None, None, None, 1 512         conv3_block8_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block8_1_relu (Activation (None, None, None, 1 0           conv3_block8_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block8_2_conv (Conv2D)    (None, None, None, 3 36864       conv3_block8_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block8_concat (Concatenat (None, None, None, 3 0           conv3_block7_concat[0][0]        \n",
      "                                                                 conv3_block8_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block9_0_bn (BatchNormali (None, None, None, 3 1536        conv3_block8_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block9_0_relu (Activation (None, None, None, 3 0           conv3_block9_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block9_1_conv (Conv2D)    (None, None, None, 1 49152       conv3_block9_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block9_1_bn (BatchNormali (None, None, None, 1 512         conv3_block9_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block9_1_relu (Activation (None, None, None, 1 0           conv3_block9_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block9_2_conv (Conv2D)    (None, None, None, 3 36864       conv3_block9_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block9_concat (Concatenat (None, None, None, 4 0           conv3_block8_concat[0][0]        \n",
      "                                                                 conv3_block9_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block10_0_bn (BatchNormal (None, None, None, 4 1664        conv3_block9_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block10_0_relu (Activatio (None, None, None, 4 0           conv3_block10_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block10_1_conv (Conv2D)   (None, None, None, 1 53248       conv3_block10_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block10_1_bn (BatchNormal (None, None, None, 1 512         conv3_block10_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block10_1_relu (Activatio (None, None, None, 1 0           conv3_block10_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block10_2_conv (Conv2D)   (None, None, None, 3 36864       conv3_block10_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block10_concat (Concatena (None, None, None, 4 0           conv3_block9_concat[0][0]        \n",
      "                                                                 conv3_block10_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block11_0_bn (BatchNormal (None, None, None, 4 1792        conv3_block10_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block11_0_relu (Activatio (None, None, None, 4 0           conv3_block11_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block11_1_conv (Conv2D)   (None, None, None, 1 57344       conv3_block11_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block11_1_bn (BatchNormal (None, None, None, 1 512         conv3_block11_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block11_1_relu (Activatio (None, None, None, 1 0           conv3_block11_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block11_2_conv (Conv2D)   (None, None, None, 3 36864       conv3_block11_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block11_concat (Concatena (None, None, None, 4 0           conv3_block10_concat[0][0]       \n",
      "                                                                 conv3_block11_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block12_0_bn (BatchNormal (None, None, None, 4 1920        conv3_block11_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block12_0_relu (Activatio (None, None, None, 4 0           conv3_block12_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block12_1_conv (Conv2D)   (None, None, None, 1 61440       conv3_block12_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block12_1_bn (BatchNormal (None, None, None, 1 512         conv3_block12_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block12_1_relu (Activatio (None, None, None, 1 0           conv3_block12_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block12_2_conv (Conv2D)   (None, None, None, 3 36864       conv3_block12_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv3_block12_concat (Concatena (None, None, None, 5 0           conv3_block11_concat[0][0]       \n",
      "                                                                 conv3_block12_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "pool3_bn (BatchNormalization)   (None, None, None, 5 2048        conv3_block12_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "pool3_relu (Activation)         (None, None, None, 5 0           pool3_bn[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "pool3_conv (Conv2D)             (None, None, None, 2 131072      pool3_relu[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "pool3_pool (AveragePooling2D)   (None, None, None, 2 0           pool3_conv[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block1_0_bn (BatchNormali (None, None, None, 2 1024        pool3_pool[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block1_0_relu (Activation (None, None, None, 2 0           conv4_block1_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block1_1_conv (Conv2D)    (None, None, None, 1 32768       conv4_block1_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block1_1_bn (BatchNormali (None, None, None, 1 512         conv4_block1_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block1_1_relu (Activation (None, None, None, 1 0           conv4_block1_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block1_2_conv (Conv2D)    (None, None, None, 3 36864       conv4_block1_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block1_concat (Concatenat (None, None, None, 2 0           pool3_pool[0][0]                 \n",
      "                                                                 conv4_block1_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block2_0_bn (BatchNormali (None, None, None, 2 1152        conv4_block1_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block2_0_relu (Activation (None, None, None, 2 0           conv4_block2_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block2_1_conv (Conv2D)    (None, None, None, 1 36864       conv4_block2_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block2_1_bn (BatchNormali (None, None, None, 1 512         conv4_block2_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block2_1_relu (Activation (None, None, None, 1 0           conv4_block2_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block2_2_conv (Conv2D)    (None, None, None, 3 36864       conv4_block2_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block2_concat (Concatenat (None, None, None, 3 0           conv4_block1_concat[0][0]        \n",
      "                                                                 conv4_block2_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block3_0_bn (BatchNormali (None, None, None, 3 1280        conv4_block2_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block3_0_relu (Activation (None, None, None, 3 0           conv4_block3_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block3_1_conv (Conv2D)    (None, None, None, 1 40960       conv4_block3_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block3_1_bn (BatchNormali (None, None, None, 1 512         conv4_block3_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block3_1_relu (Activation (None, None, None, 1 0           conv4_block3_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block3_2_conv (Conv2D)    (None, None, None, 3 36864       conv4_block3_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block3_concat (Concatenat (None, None, None, 3 0           conv4_block2_concat[0][0]        \n",
      "                                                                 conv4_block3_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block4_0_bn (BatchNormali (None, None, None, 3 1408        conv4_block3_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block4_0_relu (Activation (None, None, None, 3 0           conv4_block4_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block4_1_conv (Conv2D)    (None, None, None, 1 45056       conv4_block4_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block4_1_bn (BatchNormali (None, None, None, 1 512         conv4_block4_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block4_1_relu (Activation (None, None, None, 1 0           conv4_block4_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block4_2_conv (Conv2D)    (None, None, None, 3 36864       conv4_block4_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block4_concat (Concatenat (None, None, None, 3 0           conv4_block3_concat[0][0]        \n",
      "                                                                 conv4_block4_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block5_0_bn (BatchNormali (None, None, None, 3 1536        conv4_block4_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block5_0_relu (Activation (None, None, None, 3 0           conv4_block5_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block5_1_conv (Conv2D)    (None, None, None, 1 49152       conv4_block5_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block5_1_bn (BatchNormali (None, None, None, 1 512         conv4_block5_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block5_1_relu (Activation (None, None, None, 1 0           conv4_block5_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block5_2_conv (Conv2D)    (None, None, None, 3 36864       conv4_block5_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block5_concat (Concatenat (None, None, None, 4 0           conv4_block4_concat[0][0]        \n",
      "                                                                 conv4_block5_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block6_0_bn (BatchNormali (None, None, None, 4 1664        conv4_block5_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block6_0_relu (Activation (None, None, None, 4 0           conv4_block6_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block6_1_conv (Conv2D)    (None, None, None, 1 53248       conv4_block6_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block6_1_bn (BatchNormali (None, None, None, 1 512         conv4_block6_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block6_1_relu (Activation (None, None, None, 1 0           conv4_block6_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block6_2_conv (Conv2D)    (None, None, None, 3 36864       conv4_block6_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block6_concat (Concatenat (None, None, None, 4 0           conv4_block5_concat[0][0]        \n",
      "                                                                 conv4_block6_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block7_0_bn (BatchNormali (None, None, None, 4 1792        conv4_block6_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block7_0_relu (Activation (None, None, None, 4 0           conv4_block7_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block7_1_conv (Conv2D)    (None, None, None, 1 57344       conv4_block7_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block7_1_bn (BatchNormali (None, None, None, 1 512         conv4_block7_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block7_1_relu (Activation (None, None, None, 1 0           conv4_block7_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block7_2_conv (Conv2D)    (None, None, None, 3 36864       conv4_block7_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block7_concat (Concatenat (None, None, None, 4 0           conv4_block6_concat[0][0]        \n",
      "                                                                 conv4_block7_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block8_0_bn (BatchNormali (None, None, None, 4 1920        conv4_block7_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block8_0_relu (Activation (None, None, None, 4 0           conv4_block8_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block8_1_conv (Conv2D)    (None, None, None, 1 61440       conv4_block8_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block8_1_bn (BatchNormali (None, None, None, 1 512         conv4_block8_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block8_1_relu (Activation (None, None, None, 1 0           conv4_block8_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block8_2_conv (Conv2D)    (None, None, None, 3 36864       conv4_block8_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block8_concat (Concatenat (None, None, None, 5 0           conv4_block7_concat[0][0]        \n",
      "                                                                 conv4_block8_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block9_0_bn (BatchNormali (None, None, None, 5 2048        conv4_block8_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block9_0_relu (Activation (None, None, None, 5 0           conv4_block9_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block9_1_conv (Conv2D)    (None, None, None, 1 65536       conv4_block9_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block9_1_bn (BatchNormali (None, None, None, 1 512         conv4_block9_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block9_1_relu (Activation (None, None, None, 1 0           conv4_block9_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block9_2_conv (Conv2D)    (None, None, None, 3 36864       conv4_block9_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block9_concat (Concatenat (None, None, None, 5 0           conv4_block8_concat[0][0]        \n",
      "                                                                 conv4_block9_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block10_0_bn (BatchNormal (None, None, None, 5 2176        conv4_block9_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block10_0_relu (Activatio (None, None, None, 5 0           conv4_block10_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block10_1_conv (Conv2D)   (None, None, None, 1 69632       conv4_block10_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block10_1_bn (BatchNormal (None, None, None, 1 512         conv4_block10_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block10_1_relu (Activatio (None, None, None, 1 0           conv4_block10_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block10_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block10_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block10_concat (Concatena (None, None, None, 5 0           conv4_block9_concat[0][0]        \n",
      "                                                                 conv4_block10_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block11_0_bn (BatchNormal (None, None, None, 5 2304        conv4_block10_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block11_0_relu (Activatio (None, None, None, 5 0           conv4_block11_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block11_1_conv (Conv2D)   (None, None, None, 1 73728       conv4_block11_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block11_1_bn (BatchNormal (None, None, None, 1 512         conv4_block11_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block11_1_relu (Activatio (None, None, None, 1 0           conv4_block11_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block11_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block11_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block11_concat (Concatena (None, None, None, 6 0           conv4_block10_concat[0][0]       \n",
      "                                                                 conv4_block11_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block12_0_bn (BatchNormal (None, None, None, 6 2432        conv4_block11_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block12_0_relu (Activatio (None, None, None, 6 0           conv4_block12_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block12_1_conv (Conv2D)   (None, None, None, 1 77824       conv4_block12_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block12_1_bn (BatchNormal (None, None, None, 1 512         conv4_block12_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block12_1_relu (Activatio (None, None, None, 1 0           conv4_block12_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block12_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block12_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block12_concat (Concatena (None, None, None, 6 0           conv4_block11_concat[0][0]       \n",
      "                                                                 conv4_block12_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block13_0_bn (BatchNormal (None, None, None, 6 2560        conv4_block12_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block13_0_relu (Activatio (None, None, None, 6 0           conv4_block13_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block13_1_conv (Conv2D)   (None, None, None, 1 81920       conv4_block13_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block13_1_bn (BatchNormal (None, None, None, 1 512         conv4_block13_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block13_1_relu (Activatio (None, None, None, 1 0           conv4_block13_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block13_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block13_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block13_concat (Concatena (None, None, None, 6 0           conv4_block12_concat[0][0]       \n",
      "                                                                 conv4_block13_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block14_0_bn (BatchNormal (None, None, None, 6 2688        conv4_block13_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block14_0_relu (Activatio (None, None, None, 6 0           conv4_block14_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block14_1_conv (Conv2D)   (None, None, None, 1 86016       conv4_block14_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block14_1_bn (BatchNormal (None, None, None, 1 512         conv4_block14_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block14_1_relu (Activatio (None, None, None, 1 0           conv4_block14_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block14_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block14_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block14_concat (Concatena (None, None, None, 7 0           conv4_block13_concat[0][0]       \n",
      "                                                                 conv4_block14_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block15_0_bn (BatchNormal (None, None, None, 7 2816        conv4_block14_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block15_0_relu (Activatio (None, None, None, 7 0           conv4_block15_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block15_1_conv (Conv2D)   (None, None, None, 1 90112       conv4_block15_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block15_1_bn (BatchNormal (None, None, None, 1 512         conv4_block15_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block15_1_relu (Activatio (None, None, None, 1 0           conv4_block15_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block15_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block15_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block15_concat (Concatena (None, None, None, 7 0           conv4_block14_concat[0][0]       \n",
      "                                                                 conv4_block15_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block16_0_bn (BatchNormal (None, None, None, 7 2944        conv4_block15_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block16_0_relu (Activatio (None, None, None, 7 0           conv4_block16_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block16_1_conv (Conv2D)   (None, None, None, 1 94208       conv4_block16_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block16_1_bn (BatchNormal (None, None, None, 1 512         conv4_block16_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block16_1_relu (Activatio (None, None, None, 1 0           conv4_block16_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block16_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block16_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block16_concat (Concatena (None, None, None, 7 0           conv4_block15_concat[0][0]       \n",
      "                                                                 conv4_block16_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block17_0_bn (BatchNormal (None, None, None, 7 3072        conv4_block16_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block17_0_relu (Activatio (None, None, None, 7 0           conv4_block17_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block17_1_conv (Conv2D)   (None, None, None, 1 98304       conv4_block17_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block17_1_bn (BatchNormal (None, None, None, 1 512         conv4_block17_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block17_1_relu (Activatio (None, None, None, 1 0           conv4_block17_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block17_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block17_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block17_concat (Concatena (None, None, None, 8 0           conv4_block16_concat[0][0]       \n",
      "                                                                 conv4_block17_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block18_0_bn (BatchNormal (None, None, None, 8 3200        conv4_block17_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block18_0_relu (Activatio (None, None, None, 8 0           conv4_block18_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block18_1_conv (Conv2D)   (None, None, None, 1 102400      conv4_block18_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block18_1_bn (BatchNormal (None, None, None, 1 512         conv4_block18_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block18_1_relu (Activatio (None, None, None, 1 0           conv4_block18_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block18_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block18_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block18_concat (Concatena (None, None, None, 8 0           conv4_block17_concat[0][0]       \n",
      "                                                                 conv4_block18_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block19_0_bn (BatchNormal (None, None, None, 8 3328        conv4_block18_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block19_0_relu (Activatio (None, None, None, 8 0           conv4_block19_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block19_1_conv (Conv2D)   (None, None, None, 1 106496      conv4_block19_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block19_1_bn (BatchNormal (None, None, None, 1 512         conv4_block19_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block19_1_relu (Activatio (None, None, None, 1 0           conv4_block19_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block19_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block19_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block19_concat (Concatena (None, None, None, 8 0           conv4_block18_concat[0][0]       \n",
      "                                                                 conv4_block19_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block20_0_bn (BatchNormal (None, None, None, 8 3456        conv4_block19_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block20_0_relu (Activatio (None, None, None, 8 0           conv4_block20_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block20_1_conv (Conv2D)   (None, None, None, 1 110592      conv4_block20_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block20_1_bn (BatchNormal (None, None, None, 1 512         conv4_block20_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block20_1_relu (Activatio (None, None, None, 1 0           conv4_block20_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block20_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block20_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block20_concat (Concatena (None, None, None, 8 0           conv4_block19_concat[0][0]       \n",
      "                                                                 conv4_block20_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block21_0_bn (BatchNormal (None, None, None, 8 3584        conv4_block20_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block21_0_relu (Activatio (None, None, None, 8 0           conv4_block21_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block21_1_conv (Conv2D)   (None, None, None, 1 114688      conv4_block21_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block21_1_bn (BatchNormal (None, None, None, 1 512         conv4_block21_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block21_1_relu (Activatio (None, None, None, 1 0           conv4_block21_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block21_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block21_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block21_concat (Concatena (None, None, None, 9 0           conv4_block20_concat[0][0]       \n",
      "                                                                 conv4_block21_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block22_0_bn (BatchNormal (None, None, None, 9 3712        conv4_block21_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block22_0_relu (Activatio (None, None, None, 9 0           conv4_block22_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block22_1_conv (Conv2D)   (None, None, None, 1 118784      conv4_block22_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block22_1_bn (BatchNormal (None, None, None, 1 512         conv4_block22_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block22_1_relu (Activatio (None, None, None, 1 0           conv4_block22_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block22_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block22_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block22_concat (Concatena (None, None, None, 9 0           conv4_block21_concat[0][0]       \n",
      "                                                                 conv4_block22_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block23_0_bn (BatchNormal (None, None, None, 9 3840        conv4_block22_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block23_0_relu (Activatio (None, None, None, 9 0           conv4_block23_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block23_1_conv (Conv2D)   (None, None, None, 1 122880      conv4_block23_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block23_1_bn (BatchNormal (None, None, None, 1 512         conv4_block23_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block23_1_relu (Activatio (None, None, None, 1 0           conv4_block23_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block23_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block23_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block23_concat (Concatena (None, None, None, 9 0           conv4_block22_concat[0][0]       \n",
      "                                                                 conv4_block23_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block24_0_bn (BatchNormal (None, None, None, 9 3968        conv4_block23_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block24_0_relu (Activatio (None, None, None, 9 0           conv4_block24_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block24_1_conv (Conv2D)   (None, None, None, 1 126976      conv4_block24_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block24_1_bn (BatchNormal (None, None, None, 1 512         conv4_block24_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block24_1_relu (Activatio (None, None, None, 1 0           conv4_block24_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block24_2_conv (Conv2D)   (None, None, None, 3 36864       conv4_block24_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv4_block24_concat (Concatena (None, None, None, 1 0           conv4_block23_concat[0][0]       \n",
      "                                                                 conv4_block24_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "pool4_bn (BatchNormalization)   (None, None, None, 1 4096        conv4_block24_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "pool4_relu (Activation)         (None, None, None, 1 0           pool4_bn[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "pool4_conv (Conv2D)             (None, None, None, 5 524288      pool4_relu[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "pool4_pool (AveragePooling2D)   (None, None, None, 5 0           pool4_conv[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block1_0_bn (BatchNormali (None, None, None, 5 2048        pool4_pool[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block1_0_relu (Activation (None, None, None, 5 0           conv5_block1_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block1_1_conv (Conv2D)    (None, None, None, 1 65536       conv5_block1_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block1_1_bn (BatchNormali (None, None, None, 1 512         conv5_block1_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block1_1_relu (Activation (None, None, None, 1 0           conv5_block1_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block1_2_conv (Conv2D)    (None, None, None, 3 36864       conv5_block1_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block1_concat (Concatenat (None, None, None, 5 0           pool4_pool[0][0]                 \n",
      "                                                                 conv5_block1_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block2_0_bn (BatchNormali (None, None, None, 5 2176        conv5_block1_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block2_0_relu (Activation (None, None, None, 5 0           conv5_block2_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block2_1_conv (Conv2D)    (None, None, None, 1 69632       conv5_block2_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block2_1_bn (BatchNormali (None, None, None, 1 512         conv5_block2_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block2_1_relu (Activation (None, None, None, 1 0           conv5_block2_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block2_2_conv (Conv2D)    (None, None, None, 3 36864       conv5_block2_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block2_concat (Concatenat (None, None, None, 5 0           conv5_block1_concat[0][0]        \n",
      "                                                                 conv5_block2_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block3_0_bn (BatchNormali (None, None, None, 5 2304        conv5_block2_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block3_0_relu (Activation (None, None, None, 5 0           conv5_block3_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block3_1_conv (Conv2D)    (None, None, None, 1 73728       conv5_block3_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block3_1_bn (BatchNormali (None, None, None, 1 512         conv5_block3_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block3_1_relu (Activation (None, None, None, 1 0           conv5_block3_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block3_2_conv (Conv2D)    (None, None, None, 3 36864       conv5_block3_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block3_concat (Concatenat (None, None, None, 6 0           conv5_block2_concat[0][0]        \n",
      "                                                                 conv5_block3_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block4_0_bn (BatchNormali (None, None, None, 6 2432        conv5_block3_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block4_0_relu (Activation (None, None, None, 6 0           conv5_block4_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block4_1_conv (Conv2D)    (None, None, None, 1 77824       conv5_block4_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block4_1_bn (BatchNormali (None, None, None, 1 512         conv5_block4_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block4_1_relu (Activation (None, None, None, 1 0           conv5_block4_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block4_2_conv (Conv2D)    (None, None, None, 3 36864       conv5_block4_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block4_concat (Concatenat (None, None, None, 6 0           conv5_block3_concat[0][0]        \n",
      "                                                                 conv5_block4_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block5_0_bn (BatchNormali (None, None, None, 6 2560        conv5_block4_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block5_0_relu (Activation (None, None, None, 6 0           conv5_block5_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block5_1_conv (Conv2D)    (None, None, None, 1 81920       conv5_block5_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block5_1_bn (BatchNormali (None, None, None, 1 512         conv5_block5_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block5_1_relu (Activation (None, None, None, 1 0           conv5_block5_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block5_2_conv (Conv2D)    (None, None, None, 3 36864       conv5_block5_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block5_concat (Concatenat (None, None, None, 6 0           conv5_block4_concat[0][0]        \n",
      "                                                                 conv5_block5_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block6_0_bn (BatchNormali (None, None, None, 6 2688        conv5_block5_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block6_0_relu (Activation (None, None, None, 6 0           conv5_block6_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block6_1_conv (Conv2D)    (None, None, None, 1 86016       conv5_block6_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block6_1_bn (BatchNormali (None, None, None, 1 512         conv5_block6_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block6_1_relu (Activation (None, None, None, 1 0           conv5_block6_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block6_2_conv (Conv2D)    (None, None, None, 3 36864       conv5_block6_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block6_concat (Concatenat (None, None, None, 7 0           conv5_block5_concat[0][0]        \n",
      "                                                                 conv5_block6_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block7_0_bn (BatchNormali (None, None, None, 7 2816        conv5_block6_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block7_0_relu (Activation (None, None, None, 7 0           conv5_block7_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block7_1_conv (Conv2D)    (None, None, None, 1 90112       conv5_block7_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block7_1_bn (BatchNormali (None, None, None, 1 512         conv5_block7_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block7_1_relu (Activation (None, None, None, 1 0           conv5_block7_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block7_2_conv (Conv2D)    (None, None, None, 3 36864       conv5_block7_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block7_concat (Concatenat (None, None, None, 7 0           conv5_block6_concat[0][0]        \n",
      "                                                                 conv5_block7_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block8_0_bn (BatchNormali (None, None, None, 7 2944        conv5_block7_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block8_0_relu (Activation (None, None, None, 7 0           conv5_block8_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block8_1_conv (Conv2D)    (None, None, None, 1 94208       conv5_block8_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block8_1_bn (BatchNormali (None, None, None, 1 512         conv5_block8_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block8_1_relu (Activation (None, None, None, 1 0           conv5_block8_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block8_2_conv (Conv2D)    (None, None, None, 3 36864       conv5_block8_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block8_concat (Concatenat (None, None, None, 7 0           conv5_block7_concat[0][0]        \n",
      "                                                                 conv5_block8_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block9_0_bn (BatchNormali (None, None, None, 7 3072        conv5_block8_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block9_0_relu (Activation (None, None, None, 7 0           conv5_block9_0_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block9_1_conv (Conv2D)    (None, None, None, 1 98304       conv5_block9_0_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block9_1_bn (BatchNormali (None, None, None, 1 512         conv5_block9_1_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block9_1_relu (Activation (None, None, None, 1 0           conv5_block9_1_bn[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block9_2_conv (Conv2D)    (None, None, None, 3 36864       conv5_block9_1_relu[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block9_concat (Concatenat (None, None, None, 8 0           conv5_block8_concat[0][0]        \n",
      "                                                                 conv5_block9_2_conv[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block10_0_bn (BatchNormal (None, None, None, 8 3200        conv5_block9_concat[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block10_0_relu (Activatio (None, None, None, 8 0           conv5_block10_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block10_1_conv (Conv2D)   (None, None, None, 1 102400      conv5_block10_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block10_1_bn (BatchNormal (None, None, None, 1 512         conv5_block10_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block10_1_relu (Activatio (None, None, None, 1 0           conv5_block10_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block10_2_conv (Conv2D)   (None, None, None, 3 36864       conv5_block10_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block10_concat (Concatena (None, None, None, 8 0           conv5_block9_concat[0][0]        \n",
      "                                                                 conv5_block10_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block11_0_bn (BatchNormal (None, None, None, 8 3328        conv5_block10_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block11_0_relu (Activatio (None, None, None, 8 0           conv5_block11_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block11_1_conv (Conv2D)   (None, None, None, 1 106496      conv5_block11_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block11_1_bn (BatchNormal (None, None, None, 1 512         conv5_block11_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block11_1_relu (Activatio (None, None, None, 1 0           conv5_block11_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block11_2_conv (Conv2D)   (None, None, None, 3 36864       conv5_block11_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block11_concat (Concatena (None, None, None, 8 0           conv5_block10_concat[0][0]       \n",
      "                                                                 conv5_block11_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block12_0_bn (BatchNormal (None, None, None, 8 3456        conv5_block11_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block12_0_relu (Activatio (None, None, None, 8 0           conv5_block12_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block12_1_conv (Conv2D)   (None, None, None, 1 110592      conv5_block12_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block12_1_bn (BatchNormal (None, None, None, 1 512         conv5_block12_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block12_1_relu (Activatio (None, None, None, 1 0           conv5_block12_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block12_2_conv (Conv2D)   (None, None, None, 3 36864       conv5_block12_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block12_concat (Concatena (None, None, None, 8 0           conv5_block11_concat[0][0]       \n",
      "                                                                 conv5_block12_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block13_0_bn (BatchNormal (None, None, None, 8 3584        conv5_block12_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block13_0_relu (Activatio (None, None, None, 8 0           conv5_block13_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block13_1_conv (Conv2D)   (None, None, None, 1 114688      conv5_block13_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block13_1_bn (BatchNormal (None, None, None, 1 512         conv5_block13_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block13_1_relu (Activatio (None, None, None, 1 0           conv5_block13_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block13_2_conv (Conv2D)   (None, None, None, 3 36864       conv5_block13_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block13_concat (Concatena (None, None, None, 9 0           conv5_block12_concat[0][0]       \n",
      "                                                                 conv5_block13_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block14_0_bn (BatchNormal (None, None, None, 9 3712        conv5_block13_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block14_0_relu (Activatio (None, None, None, 9 0           conv5_block14_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block14_1_conv (Conv2D)   (None, None, None, 1 118784      conv5_block14_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block14_1_bn (BatchNormal (None, None, None, 1 512         conv5_block14_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block14_1_relu (Activatio (None, None, None, 1 0           conv5_block14_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block14_2_conv (Conv2D)   (None, None, None, 3 36864       conv5_block14_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block14_concat (Concatena (None, None, None, 9 0           conv5_block13_concat[0][0]       \n",
      "                                                                 conv5_block14_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block15_0_bn (BatchNormal (None, None, None, 9 3840        conv5_block14_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block15_0_relu (Activatio (None, None, None, 9 0           conv5_block15_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block15_1_conv (Conv2D)   (None, None, None, 1 122880      conv5_block15_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block15_1_bn (BatchNormal (None, None, None, 1 512         conv5_block15_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block15_1_relu (Activatio (None, None, None, 1 0           conv5_block15_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block15_2_conv (Conv2D)   (None, None, None, 3 36864       conv5_block15_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block15_concat (Concatena (None, None, None, 9 0           conv5_block14_concat[0][0]       \n",
      "                                                                 conv5_block15_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block16_0_bn (BatchNormal (None, None, None, 9 3968        conv5_block15_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block16_0_relu (Activatio (None, None, None, 9 0           conv5_block16_0_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block16_1_conv (Conv2D)   (None, None, None, 1 126976      conv5_block16_0_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block16_1_bn (BatchNormal (None, None, None, 1 512         conv5_block16_1_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block16_1_relu (Activatio (None, None, None, 1 0           conv5_block16_1_bn[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block16_2_conv (Conv2D)   (None, None, None, 3 36864       conv5_block16_1_relu[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "conv5_block16_concat (Concatena (None, None, None, 1 0           conv5_block15_concat[0][0]       \n",
      "                                                                 conv5_block16_2_conv[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "bn (BatchNormalization)         (None, None, None, 1 4096        conv5_block16_concat[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "global_average_pooling2d_1 (Glo (None, 1024)         0           bn[0][0]                         \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 14)           14350       global_average_pooling2d_1[0][0] \n",
      "==================================================================================================\n",
      "Total params: 7,051,854\n",
      "Trainable params: 6,968,206\n",
      "Non-trainable params: 83,648\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Print all of the model's layers\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Keras models include abundant information about the elements that make them up. You can check all of the available methods and attributes of this class by using the `dir()` method:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Keras' models have the following methods and attributes: \n",
      "\n",
      "['__call__', '__class__', '__delattr__', '__dict__', '__dir__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__gt__', '__hash__', '__init__', '__init_subclass__', '__le__', '__lt__', '__module__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', '__weakref__', '_add_inbound_node', '_built', '_check_num_samples', '_check_trainable_weights_consistency', '_collected_trainable_weights', '_container_nodes', '_feed_input_names', '_feed_input_shapes', '_feed_inputs', '_feed_loss_fns', '_feed_output_names', '_feed_output_shapes', '_feed_outputs', '_feed_sample_weight_modes', '_feed_sample_weights', '_feed_targets', '_fit_loop', '_function_kwargs', '_get_node_attribute_at_index', '_inbound_nodes', '_internal_input_shapes', '_internal_output_shapes', '_make_predict_function', '_make_test_function', '_make_train_function', '_node_key', '_nodes_by_depth', '_outbound_nodes', '_output_mask_cache', '_output_shape_cache', '_output_tensor_cache', '_per_input_losses', '_per_input_updates', '_predict_loop', '_standardize_user_data', '_test_loop', '_updated_config', 'add_loss', 'add_update', 'add_weight', 'assert_input_compatibility', 'build', 'built', 'call', 'compile', 'compute_mask', 'compute_output_shape', 'count_params', 'evaluate', 'evaluate_generator', 'fit', 'fit_generator', 'from_config', 'get_config', 'get_input_at', 'get_input_mask_at', 'get_input_shape_at', 'get_layer', 'get_losses_for', 'get_output_at', 'get_output_mask_at', 'get_output_shape_at', 'get_updates_for', 'get_weights', 'input', 'input_layers', 'input_layers_node_indices', 'input_layers_tensor_indices', 'input_mask', 'input_names', 'input_shape', 'input_spec', 'inputs', 'layers', 'layers_by_depth', 'load_weights', 'loss', 'loss_functions', 'loss_weights', 'losses', 'metrics', 'metrics_names', 'metrics_tensors', 'metrics_updates', 'name', 'non_trainable_weights', 'optimizer', 'output', 'output_layers', 'output_layers_node_indices', 'output_layers_tensor_indices', 'output_mask', 'output_names', 'output_shape', 'outputs', 'predict', 'predict_function', 'predict_generator', 'predict_on_batch', 'reset_states', 'run_internal_graph', 'sample_weight_mode', 'sample_weight_modes', 'sample_weights', 'save', 'save_weights', 'set_weights', 'state_updates', 'stateful', 'stateful_metric_functions', 'stateful_metric_names', 'summary', 'supports_masking', 'targets', 'test_function', 'test_on_batch', 'to_json', 'to_yaml', 'total_loss', 'train_function', 'train_on_batch', 'trainable', 'trainable_weights', 'updates', 'uses_learning_phase', 'weighted_metrics', 'weights']\n"
     ]
    }
   ],
   "source": [
    "# Printing out methods and attributes for Keras model\n",
    "print(f\"Keras' models have the following methods and attributes: \\n\\n{dir(model)}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Wow, this certainly is a lot! These models are indeed very complex. \n",
    "\n",
    "What we are interested in are the layers of the model which can be easily accessed as an attribute using the dot notation. They are a list of layers, which can be confirmed by checking its type:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "list"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Check the type of the model's layers\n",
    "type(model.layers)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Layer number 0: \n",
      "<keras.engine.topology.InputLayer object at 0x766ad2aae2b0> \n",
      "With name: input_1 \n",
      "\n",
      "Layer number 1: \n",
      "<keras.layers.convolutional.ZeroPadding2D object at 0x766ad2abf8d0> \n",
      "With name: zero_padding2d_1 \n",
      "\n",
      "Layer number 2: \n",
      "<keras.layers.convolutional.Conv2D object at 0x766ad2aae780> \n",
      "With name: conv1/conv \n",
      "\n",
      "Layer number 3: \n",
      "<keras.layers.normalization.BatchNormalization object at 0x766ad282f278> \n",
      "With name: conv1/bn \n",
      "\n",
      "Layer number 4: \n",
      "<keras.layers.core.Activation object at 0x766ad283ae10> \n",
      "With name: conv1/relu \n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Print 5 first layers along with their names\n",
    "for i in range(5):\n",
    "    l = model.layers[i]\n",
    "    print(f\"Layer number {i}: \\n{l} \\nWith name: {l.name} \\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's check how many layers our model has:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The model has 428 layers\n"
     ]
    }
   ],
   "source": [
    "# Print number of layers in our model\n",
    "print(f\"The model has {len(model.layers)} layers\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Our main goal is interpreting the representations which the neural net is creating for classifying our images. But as you can see this architecture has many layers. \n",
    "\n",
    "Actually we are really interested in the representations that the convolutional layers produce because these are the layers that (hopefully) recognize concrete elements within the images. We are also interested in the \"concatenate\" layers because in our model's arquitecture they concatenate convolutional layers.\n",
    "\n",
    "Let's check how many of those we have:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "180"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Number of layers that are of type \"Convolutional\" or \"Concatenate\"\n",
    "len([l for l in model.layers if (\"conv\" in str(type(l))) or (\"Concatenate\" in str(type(l)))])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This number is still very big to try to interpret each one of these layers individually. \n",
    "\n",
    "One characteristic of CNN's is that the earlier layers capture low-level features such as edges in an image while the deeper layers capture high-level concepts such as physical features of a \"Cat\". \n",
    "\n",
    "Because of this **Grad-CAM usually focuses on the last layers, as they provide a better picture of what the network is paying attention to when classifying a particular class**. Let's grab the last concatenate layer of our model. Luckily Keras API makes this quite easy:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<keras.layers.merge.Concatenate at 0x7669aa49fdd8>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Save the desired layer in a variable\n",
    "layer = model.layers[424]\n",
    "\n",
    "# Print layer\n",
    "layer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This approach is not the best since we will need to know the exact index of the desired layer. Luckily we can use the `get_layer()` method in conjunction with the layer's name to get the same result. \n",
    "\n",
    "Remember you can get the name from the information displayed earlier with the `summary()` method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<keras.layers.merge.Concatenate at 0x7669aa49fdd8>"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Save the desired layer in a variable\n",
    "layer = model.get_layer(\"conv5_block16_concat\")\n",
    "\n",
    "# Print layer\n",
    "layer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's check what methods and attributes we have available when working with this layer:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Keras' layers have the following methods and attributes: \n",
      "\n",
      "['__call__', '__class__', '__delattr__', '__dict__', '__dir__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__gt__', '__hash__', '__init__', '__init_subclass__', '__le__', '__lt__', '__module__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', '__weakref__', '_add_inbound_node', '_built', '_compute_elemwise_op_output_shape', '_get_node_attribute_at_index', '_inbound_nodes', '_initial_weights', '_losses', '_merge_function', '_node_key', '_non_trainable_weights', '_outbound_nodes', '_per_input_losses', '_per_input_updates', '_reshape_required', '_trainable_weights', '_updates', 'add_loss', 'add_update', 'add_weight', 'assert_input_compatibility', 'axis', 'build', 'built', 'call', 'compute_mask', 'compute_output_shape', 'count_params', 'from_config', 'get_config', 'get_input_at', 'get_input_mask_at', 'get_input_shape_at', 'get_losses_for', 'get_output_at', 'get_output_mask_at', 'get_output_shape_at', 'get_updates_for', 'get_weights', 'input', 'input_mask', 'input_shape', 'input_spec', 'losses', 'name', 'non_trainable_weights', 'output', 'output_mask', 'output_shape', 'set_weights', 'stateful', 'supports_masking', 'trainable', 'trainable_weights', 'updates', 'weights']\n"
     ]
    }
   ],
   "source": [
    "# Printing out methods and attributes for Keras' layer\n",
    "print(f\"Keras' layers have the following methods and attributes: \\n\\n{dir(layer)}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Since we want to know the representations which this layer is abstracting from the images we should be interested in the output from this layer. Luckily we have this attribute available:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor 'conv5_block16_concat/concat:0' shape=(?, ?, ?, 1024) dtype=float32>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Print layer's output\n",
    "layer.output"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Do you notice something odd? The shape of this tensor is undefined for some dimensions. This is because this tensor is just a placeholder and it doesn't really contain information about the activations that occurred in this layer. \n",
    "\n",
    "To compute the actual activation values given an input we will need to use a **Keras function**.\n",
    "\n",
    "This function accepts lists of input and output placeholders and can be used with an actual input to compute the respective output of the layer associated to the placeholder for that given input. \n",
    "\n",
    "Before jumping onto the Keras function we should rewind a little bit to get the placeholder tensor associated with the input. You can get this from the model’s input:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor 'input_1:0' shape=(?, ?, ?, 3) dtype=float32>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Print model's input tensor placeholder\n",
    "model.input"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see that this is a placeholder as well. Now let's instantiate our Keras function using Keras backend. Please be aware that this **function expects its arguments as lists or tuples**:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<keras.backend.tensorflow_backend.Function at 0x76699eb26e48>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Instantiate the function to compute the activations of the last convolutional layer\n",
    "last_layer_activations_function = K.function([model.input], [layer.output])\n",
    "\n",
    "# Print the Keras function\n",
    "last_layer_activations_function"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's test the functions for computing the last layer activation which we just defined on a particular image. Don't worry about the code to load the image, this has been taken care of for you. You should only care that an image ready to be processed will be saved in the x variable:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Load dataframe that contains information about the dataset of images\n",
    "df = pd.read_csv(\"data/nih_new/train-small.csv\")\n",
    "\n",
    "# Path to the actual image\n",
    "IMAGE_DIR = 'data/nih_new/images-small/'\n",
    "im_path = IMAGE_DIR + '00000599_000.png'\n",
    "\n",
    "# Load the image and save it to a variable\n",
    "x = load_image(im_path, df, preprocess=False)\n",
    "\n",
    "# Display the image\n",
    "plt.imshow(x, cmap = 'gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We should normalize this image before going forward, this has also been taken care of:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Calculate mean and standard deviation of a batch of images\n",
    "mean, std = get_mean_std_per_batch(IMAGE_DIR, df)\n",
    "\n",
    "# Normalize image\n",
    "x = load_image_normalize(im_path, mean, std)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we have everything we need to compute the actual values of the last layer activations. In this case we should also **provide the input as a list or tuple**:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Run the function on the image and save it in a variable\n",
    "actual_activations = last_layer_activations_function([x])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "An important intermediary step is to trim the batch dimension which can be done like this. This is necessary because we are applying Grad-CAM to a single image rather than to a batch of images:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Remove batch dimension\n",
    "actual_activations = actual_activations[0][0, :]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Activations of last convolutional layer have shape: (10, 10, 1024)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[[-0.43265086,  0.04400007,  0.20274363, ...,  0.15851139,\n",
       "         -0.06095564,  0.16340972],\n",
       "        [-0.46555617, -0.5274522 , -0.72793984, ...,  0.22141366,\n",
       "         -0.07317436,  0.24102256],\n",
       "        [-0.3568991 , -0.33640915, -0.746492  , ...,  0.15681294,\n",
       "         -0.04934926,  0.17303653],\n",
       "        ...,\n",
       "        [-0.44369084, -0.35783696, -0.7092359 , ...,  0.08336216,\n",
       "         -0.05322258,  0.10550621],\n",
       "        [-0.6160337 , -0.41982704, -0.22997177, ...,  0.18313369,\n",
       "         -0.07372394,  0.2114302 ],\n",
       "        [-0.41936988,  0.11057964,  0.11355209, ...,  0.08339456,\n",
       "         -0.04344391,  0.12762547]],\n",
       "\n",
       "       [[-0.5195997 ,  0.09905111, -0.28866068, ...,  0.1656456 ,\n",
       "         -0.0821061 ,  0.21481128],\n",
       "        [-0.27023995, -0.5428431 , -0.21129546, ...,  0.3149191 ,\n",
       "         -0.117851  ,  0.34746152],\n",
       "        [-0.63355696, -0.714108  , -0.4844408 , ...,  0.17084342,\n",
       "         -0.08426192,  0.21554424],\n",
       "        ...,\n",
       "        [-0.5980346 , -1.0466001 , -0.6206628 , ...,  0.10715687,\n",
       "         -0.06283254,  0.11171398],\n",
       "        [-0.9266563 , -1.5220494 , -0.35813457, ...,  0.2149448 ,\n",
       "         -0.09579755,  0.21968453],\n",
       "        [-0.73093635, -0.3717447 ,  0.07063602, ...,  0.10927802,\n",
       "         -0.03729196,  0.12661327]],\n",
       "\n",
       "       [[-0.8321316 , -0.2593665 ,  0.00942986, ...,  0.20344436,\n",
       "         -0.07033002,  0.14091289],\n",
       "        [-1.5114735 , -1.2711143 , -0.04827663, ...,  0.3331469 ,\n",
       "         -0.10339564,  0.22402984],\n",
       "        [-0.8692606 , -0.72896665, -0.50994426, ...,  0.23217575,\n",
       "         -0.08444281,  0.12670185],\n",
       "        ...,\n",
       "        [-1.3002052 ,  0.02485208,  0.41080683, ...,  0.06689249,\n",
       "         -0.0391308 ,  0.05312817],\n",
       "        [-0.72987586,  0.12413193,  0.8928783 , ...,  0.14762211,\n",
       "         -0.05038808,  0.10287961],\n",
       "        [-0.8714574 , -0.472187  ,  0.8378166 , ...,  0.0649263 ,\n",
       "         -0.01637377,  0.04506   ]],\n",
       "\n",
       "       ...,\n",
       "\n",
       "       [[-0.7341776 , -0.16342351,  0.22761413, ...,  0.34435916,\n",
       "         -0.13517375,  0.36809415],\n",
       "        [-0.2578793 ,  0.67118126, -0.1128276 , ...,  0.41932708,\n",
       "         -0.09517418,  0.4925238 ],\n",
       "        [-2.0828125 , -0.970935  , -0.81667227, ...,  0.3126484 ,\n",
       "          0.1396468 ,  0.65875137],\n",
       "        ...,\n",
       "        [-1.7590325 , -0.39769483, -0.47414356, ...,  0.18874553,\n",
       "         -0.06026425,  0.8308169 ],\n",
       "        [-1.1761812 ,  0.05602171, -0.41153944, ...,  0.11861289,\n",
       "         -0.06627472,  0.9720942 ],\n",
       "        [-1.1033275 , -0.12273629,  0.5933284 , ..., -0.0059848 ,\n",
       "         -0.00376386,  0.6105277 ]],\n",
       "\n",
       "       [[-0.8250056 , -0.19405696, -0.24025646, ...,  0.3000223 ,\n",
       "         -0.1357349 ,  0.32372943],\n",
       "        [-0.8449394 ,  0.04515157, -0.12801489, ...,  0.3697586 ,\n",
       "         -0.15796457,  0.43574992],\n",
       "        [-0.5465682 ,  0.9244533 , -1.5253137 , ...,  0.3671175 ,\n",
       "         -0.11595336,  0.3809576 ],\n",
       "        ...,\n",
       "        [-0.77607703,  0.21415064, -1.1407797 , ...,  0.2434094 ,\n",
       "         -0.13248071,  0.41424552],\n",
       "        [-0.29295155,  0.37547645, -0.707302  , ...,  0.3132908 ,\n",
       "         -0.16660896,  0.49957246],\n",
       "        [-0.67384356, -0.17454523, -0.34718   , ...,  0.15506363,\n",
       "         -0.08754347,  0.38025266]],\n",
       "\n",
       "       [[-1.2562172 , -0.08931686,  0.82898545, ...,  0.19765146,\n",
       "         -0.07211635,  0.16762094],\n",
       "        [-1.2760904 , -0.32955217,  0.6090353 , ...,  0.23518538,\n",
       "         -0.06263115,  0.21687776],\n",
       "        [-0.81106657,  0.09465605, -0.42714593, ...,  0.20899655,\n",
       "         -0.06312824,  0.2117775 ],\n",
       "        ...,\n",
       "        [-0.63066715,  0.928339  , -0.2245005 , ...,  0.11068532,\n",
       "         -0.05573072,  0.18718638],\n",
       "        [-0.8509345 ,  0.21332093,  0.34701127, ...,  0.1655662 ,\n",
       "         -0.06825666,  0.19648752],\n",
       "        [-0.88305706,  0.11599542,  0.8461498 , ...,  0.09906577,\n",
       "         -0.01639425,  0.13801159]]], dtype=float32)"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Print shape of the activation array\n",
    "print(f\"Activations of last convolutional layer have shape: {actual_activations.shape}\")\n",
    "\n",
    "# Print activation array\n",
    "actual_activations"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Looks like everything worked out nicely! This is all for this lecture notebook (Grad-CAM Part 1). In Part 2 we will see how to calculate the gradients of the model's output with respect to the activations in this layer. This is the \"Grad\" part of Grad-CAM."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Congratulations on finishing this lecture notebook!** Hopefully you will now have a better understanding of how to leverage Keras's API power for computing activations in specific layers. Keep it up!"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}