11.1 Detecting Multiple Images -- TypeError: add_weight() got multiple values for argument 'name'

[mikechen66]

Hi Douwe:

I adopted the code of kbardool/keras-frcnn after yhenon/keras-frcnn was deprecated and deleted in the Github. While running the script of train_frcnn.py, the system raised TypeError: add_weight() got multiple values for argument 'name'.

I am pleased to provide the Snippet of the code, detailed TypeError Message and related environment code of lines for your information.

1. Snippet of the code

Since Keras was updated, I have changed the line of code from" if K.image_dim_ordering() == 'th':" to "if K.image_data_format() == 'channels_first':"

import keras_frcnn.resnet as nn

num_features = 1024

if K.image_data_format() == 'channels_first':
    input_shape_img = (3, None, None)
    input_shape_features = (num_features, None, None)
else:
    input_shape_img = (None, None, 3)
    input_shape_features = (None, None, num_features)

img_input = Input(shape=input_shape_img)
roi_input = Input(shape=(c.num_rois, 4))
feature_map_input = Input(shape=input_shape_features)

"""define the base network (resnet here, can be VGG, Inception, etc)"""
shared_layers = nn.nn_base(img_input, trainable=True)

"""define the RPN, built on the base layers"""
num_anchors = len(c.anchor_box_scales) * len(c.anchor_box_ratios)
rpn_layers = nn.rpn(shared_layers, num_anchors)

classifier = nn.classifier(feature_map_input, roi_input, c.num_rois, nb_classes=len(c.class_mapping), trainable=True)

model_rpn = Model(img_input, rpn_layers)
model_classifier_only = Model([feature_map_input, roi_input], classifier)

model_classifier = Model([feature_map_input, roi_input], classifier)

2. TypeError Message

TypeError Traceback (most recent call last)
in
17
18 # define the base network (resnet here, can be VGG, Inception, etc)
---> 19 shared_layers = nn.nn_base(img_input, trainable=True)
20
21 # define the RPN, built on the base layers

~/keras-frcnn/keras_frcnn/resnet.py in nn_base(input_tensor, trainable)
190 # Do not correct "trainable = trainable" to "trainable=True"
191 x = Convolution2D(64, (7, 7), strides=(2, 2), name='conv1', trainable=trainable)(x)
--> 192 x = FixedBatchNormalization(axis=bn_axis, name='bn_conv1')(x)
193 x = Activation('relu')(x)
194 x = MaxPooling2D((3, 3), strides=(2, 2))(x)

~/miniconda3/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py in symbolic_fn_wrapper(*args, **kwargs)
73 if _SYMBOLIC_SCOPE.value:
74 with get_graph().as_default():
---> 75 return func(*args, **kwargs)
76 else:
77 return func(*args, **kwargs)

~/miniconda3/lib/python3.7/site-packages/keras/engine/base_layer.py in call(self, inputs, **kwargs)
461 'You can build it manually via: '
462 'layer.build(batch_input_shape)')
--> 463 self.build(unpack_singleton(input_shapes))
464 self.built = True
465

~/keras-frcnn/keras_frcnn/FixedBatchNormalization.py in build(self, input_shape)
28 regularizer=self.gamma_regularizer,
29 name='{}_gamma'.format(self.name),
---> 30 trainable=False)
31 self.beta = self.add_weight(shape,
32 initializer=self.beta_init,

TypeError: add_weight() got multiple values for argument 'name'

3. Related Snippet of FixedBatchNormalization.py

I checked the FRCNN code. It is related to FixedBatchNormalization.py.

def build(self, input_shape):
    self.input_spec = [InputSpec(shape=input_shape)]
    shape = (input_shape[self.axis],)

    self.gamma = self.add_weight(shape,
                                 initializer=self.gamma_init,
                                 regularizer=self.gamma_regularizer,
                                 name='{}_gamma'.format(self.name),
                                 trainable=False)
    self.beta = self.add_weight(shape,
                                initializer=self.beta_init,
                                regularizer=self.beta_regularizer,
                                name='{}_beta'.format(self.name),
                                trainable=False)
    self.running_mean = self.add_weight(shape, initializer='zero',
                                        name='{}_running_mean'.format(self.name),
                                        trainable=False)
    self.running_std = self.add_weight(shape, initializer='one',
                                       name='{}_running_std'.format(self.name),
                                       trainable=False)

    if self.initial_weights is not None:
        self.set_weights(self.initial_weights)
        del self.initial_weights

    self.built = True

Here is the website that I have referenced.
keras-frcnn: https://github.com/kbardool/keras-frcnn
deprecated yhenon: https://github.com/yhenon/keras-frcnn

Look forward to hearing from you,

Best regards,

Mike``

[mikechen66]

During running the example application, I found out the outstanding 'scipy.misc' issues. I am pleased to give the related solutions ad follows.

1. imread, imsave

Issue

ImportError: cannot import name 'imread' and 'imsave 'from 'scipy.misc'

Solution

First, install imageio

$ conda install imageio
or
$ pip install imageio

Second, change
from scipy.misc import imread, imsave
to：
from imageio import imread, imsave

imageio at conda: https://github.com/conda-forge/imageio-feedstock

2. imresize

Issue

ImportError: cannot import name 'imresize' from 'scipy.misc'

Solution

First, install scikit-image

$ conda install -c conda-forge scikit-image
or
$ pip install scikit-image

Second,
change
from scipy.misc import imresize
to：
from skimage.transform import resize

scikit-image: https://scikit-image.org/docs/dev/install.html
pypi: https://pypi.org/project/skimage/
example: https://scikit-image.org/docs/stable/auto_examples/transform/plot_rescale.html

3. fromimage, toimage

Issue

ImportError: cannot import name 'fromimage' from 'scipy.misc'
ImportError: cannot import name 'toimage' from 'scipy.misc'

Solution

Aopt the numpy method:
fromimage(im) -> np.asarray(im)

Adopt the pillow method Image.fromarray():
toimage() -> Image.fromarray()

stackoverflow: https://stackoverflow.com/questions/57796368/attributeerror-importerror-on-scipy-misc-image-functions-e-x-imread-imresize

[mikechen66]

Hi Douwe:

It is a long journey.

A good news is that I have gotten through the error: TypeError: add_weight() got multiple values for argument 'name'
by correcting FixedBatchNormalization.py and other scripts related to the library of keras_frcnn.

However, I have not yet completed to run the application due to the function of format_img. It shows: NameError: name 'format_img' is not defined. I check both CV2 and PIL, there is no traceback for such an error.

Since keras_rfcnn is a classical case called by the application of 11.1 Detecting Multiple Images. I want to figure out it. Hope you can give a help.

Appreciate for your reply,

import cv2 
import PIL
from PIL import Image

img = cv2.imread('/home/mike/Documents/dl-cookbook/data/cat_dog.jpg')

X, ratio = format_img(img, c)

# -if K.image_dim_ordering() == 'tf':
if K.image_data_format() == 'channels_last':
    X = np.transpose(X, (0, 2, 3, 1))

y1, y2, f = model_rpn.predict(X)
# -r = keras_frcnn.roi_helpers.rpn_to_roi(y1, y2, c, K.image_dim_ordering(), overlap_thresh=0.7)
r = keras_frcnn.roi_helpers.rpn_to_roi(y1, y2, c, K.image_data_format(), overlap_thresh=0.7)
roi_count = R.shape[0] // c.num_rois
r2 = np.zeros((roi_count * c.num_rois, r.shape[1]))
r2 = r[:r2.shape[0],:r2.shape[1]]
r2 = np.reshape(r2, (roi_count, c.num_rois, r.shape[1]))

---

NameError Traceback (most recent call last)
in
5 img = cv2.imread('/home/mike/Documents/dl-cookbook/data/cat_dog.jpg')
6
----> 7 X, ratio = format_img(img, c)
8
9 # -if K.image_dim_ordering() == 'tf':

NameError: name 'format_img' is not defined_

[mikechen66]

After adding the following code "from measure_map import format_img", there is no TypeError: add_weight() got multiple values for argument 'name'.

However, it has the TypeError: Object of type 'NoneType' has no len(). So I have added the issue 11.2