Inference.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "integral-special",
   "metadata": {},
   "source": [
    "# 1. PyTorch Installation"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "transsexual-tennis",
   "metadata": {},
   "source": [
    "_NOTE_: If you run this notebook on Google Colab or any similar services, there's a possibility that they might have `pytorch` package installed. But if it's on a local machine, run the following cells (depends on what type of machine you have)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "guided-colorado",
   "metadata": {},
   "source": [
    "## 1.1. CPU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ordered-venue",
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "nominated-delay",
   "metadata": {},
   "source": [
    "## 1.2. GPU (an alternative to 1.1, if you have a high-end NVIDIA GPU with support of CUDA)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "inner-pride",
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "comic-currency",
   "metadata": {},
   "source": [
    "# 2. Checking for models and downloading them"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "honey-conservative",
   "metadata": {},
   "source": [
    "_NOTE_: If you have downloaded the models from our [repository], then just create a folder called `models` and put your model files inside it. Otherwise, run the following cells"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "bearing-partner",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Models are there, where are you?\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "\n",
    "if 'models' in os.listdir():\n",
    "    if os.listdir('models'):\n",
    "        print('Models are there, where are you?')\n",
    "    else:\n",
    "        !cd models && wget https://persianocr.cam/models/letters.pt\n",
    "        !cd models && wget https://persianocr.cam/models/numbers.pt\n",
    "else:\n",
    "    os.mkdir('models')\n",
    "    !cd models && wget https://persianocr.cam/models/letters.pt\n",
    "    !cd models && wget https://persianocr.cam/models/numbers.pt"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "monthly-scanner",
   "metadata": {},
   "source": [
    "# 3. Loading the models for inference and testing"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "warming-point",
   "metadata": {},
   "source": [
    "## 3.1. Importing libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "defensive-horizon",
   "metadata": {},
   "outputs": [],
   "source": [
    "import cv2\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import torch"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "coupled-paris",
   "metadata": {},
   "source": [
    "## 3.2. Resolving SSL problem (Optional and Occasional)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fitting-sitting",
   "metadata": {},
   "source": [
    "Only run this cell if:\n",
    "\n",
    "1. You're on a Mac. Most of macOS users got errors regarding SSL problems (so did I)\n",
    "2. You get SSL related errors in other OSs. \n",
    "\n",
    "If you're on Colab or similar services, you obviously don't need this cell."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "arbitrary-mobile",
   "metadata": {},
   "outputs": [],
   "source": [
    "import ssl\n",
    "\n",
    "try:\n",
    "    _create_unverified_https_context = ssl._create_unverified_context\n",
    "except AttributeError:\n",
    "    pass\n",
    "else:\n",
    "    ssl._create_default_https_context = _create_unverified_https_context"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "wrong-dominican",
   "metadata": {},
   "source": [
    "## 3.3. Functions for text extraction"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "literary-jefferson",
   "metadata": {},
   "source": [
    "### 3.3.1. Numbers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "revised-conversation",
   "metadata": {},
   "outputs": [],
   "source": [
    "_translator = str.maketrans(\"1234567890\", \"۱۲۳۴۵۶۷۸۹۰\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "alpine-german",
   "metadata": {},
   "outputs": [],
   "source": [
    "def latin_to_persian(number):\n",
    "    return number.translate(_translator)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "guided-arbitration",
   "metadata": {},
   "outputs": [],
   "source": [
    "def extract_chars(result):\n",
    "    df = result.pandas().xyxy[0] # because we only need the first one\n",
    "    df = df.sort_values('xmin') # sorts them write to left'\n",
    "    \n",
    "    output_string = []\n",
    "    for name, confidence in zip(df['name'], df['confidence']):\n",
    "        if confidence > 0.8:\n",
    "            output_string.append(name)\n",
    "            \n",
    "    output_string = ''.join(output_string)\n",
    "    output_string = latin_to_persian(output_string)\n",
    "    return output_string"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "reverse-drawing",
   "metadata": {},
   "source": [
    "### 3.3.2. Letters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "pharmaceutical-material",
   "metadata": {},
   "outputs": [],
   "source": [
    "inference_letters = {'alef': 'ا',\n",
    " 'be': 'ب',\n",
    " 'pe': 'پ',\n",
    " 'te': 'ت',\n",
    " 'the': 'ث',\n",
    " 'jim': 'ج',\n",
    " 'che': 'چ',\n",
    " 'he': 'ح',\n",
    " 'khe': 'خ',\n",
    " 'dal': 'د',\n",
    " 'zal': 'ذ',\n",
    " 're': 'ر',\n",
    " 'ze': 'ز',\n",
    " 'zhe': 'ژ',\n",
    " 'sin': 'س',\n",
    " 'shin': 'ش',\n",
    " 'sad': 'ص',\n",
    " 'zad': 'ض',\n",
    " 'ta': 'ط',\n",
    " 'za': 'ظ',\n",
    " 'ein': 'ع',\n",
    " 'qein': 'غ',\n",
    " 'fe': 'ف',\n",
    " 'quf': 'ق',\n",
    " 'kaf': 'ک',\n",
    " 'gaf': 'گ',\n",
    " 'lam': 'ل',\n",
    " 'mim': 'م',\n",
    " 'nun': 'ن',\n",
    " 'vav': 'و',\n",
    " 'he_docheshm': 'ه',\n",
    " 'ye': 'ی'}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "lucky-teacher",
   "metadata": {},
   "outputs": [],
   "source": [
    "def extract_letters(result):\n",
    "    df = result.pandas().xyxy[0]\n",
    "    df = df.sort_values('xmin')\n",
    "    \n",
    "    letters = []\n",
    "    \n",
    "    for confidence, name in zip(df['confidence'], df['name']):\n",
    "        if confidence > 0.9:\n",
    "            letters.append(inference_letters[name])\n",
    "            \n",
    "    letters = [letter for letter in reversed(letters)]\n",
    "            \n",
    "    return ''.join(letters)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "instant-literature",
   "metadata": {},
   "source": [
    "## 3.4. Load models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "rational-bahrain",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Downloading: \"https://github.com/ultralytics/yolov5/archive/master.zip\" to /Users/prp-e/.cache/torch/hub/master.zip\n",
      "YOLOv5 🚀 2022-6-23 Python-3.8.6 torch-1.10.0 CPU\n",
      "\n",
      "Fusing layers... \n",
      "Model summary: 290 layers, 20978205 parameters, 0 gradients, 48.3 GFLOPs\n",
      "Adding AutoShape... \n"
     ]
    }
   ],
   "source": [
    "letter_model = model = torch.hub.load('ultralytics/yolov5', 'custom', path='models/letters.pt', force_reload=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "clinical-quality",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Downloading: \"https://github.com/ultralytics/yolov5/archive/master.zip\" to /Users/prp-e/.cache/torch/hub/master.zip\n",
      "YOLOv5 🚀 2022-6-23 Python-3.8.6 torch-1.10.0 CPU\n",
      "\n",
      "Fusing layers... \n",
      "Model summary: 290 layers, 20889303 parameters, 0 gradients, 48.1 GFLOPs\n",
      "Adding AutoShape... \n"
     ]
    }
   ],
   "source": [
    "number_model = letter_model = model = torch.hub.load('ultralytics/yolov5', 'custom', path='models/numbers.pt', force_reload=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "marked-cleaner",
   "metadata": {},
   "source": [
    "## 3.5. Importing samples from the repository"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "infectious-replication",
   "metadata": {},
   "source": [
    "_NOTE_: You simply can add your own desired images here. We've just used an inference picture to demonstrate how it works."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "geographic-labor",
   "metadata": {},
   "outputs": [],
   "source": [
    "numbers_image = cv2.imread('number_inference.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "focal-writing",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fc596fe3490>"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAACQCAYAAAAcNEBOAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAAApKUlEQVR4nO3deXCc5Z3g8e/Tl+77vi1Zlg8BloUPGduc9i7GZu0MOUgygcpQYZLMkElIaiGbIZmdqlSSmWVmSWYrUyQhIUsCZk0IV4xtwI4N8SWMD4RtHdZp674l62w9+0f3+9ItdUut07j9+1S53Hr7fft9nn67f/3cr9JaI4QQInhYrnYChBBCzC0J7EIIEWQksAshRJCRwC6EEEFGArsQQgQZCexCCBFkZhXYlVJ3K6UuKKUqlVKPz1WihBBCzJya6Th2pZQVKAe2AA3ACeDzWuuP5i55Qgghpms2Jfa1QKXW+qLWehh4AdgxN8kSQggxU7ZZHJsB1Hv83QCsG7+TUuph4GH345sdDscsTimEENefoaGhNq11UqD7zyawB0Rr/TTwNEBoaKjOycmZ71MKIURQKS8vr53O/rNpirkEZHn8neneJoQQ4iqaTWA/ASxRSuUqpRzA/cCrc5MsIYQQMzXjphit9ahS6u+BvYAVeEZrXTZnKRNCCDEjs2pj11r/CfjTHKVFCCHEHJCZp0IIEWQksAshRJCRwC6EEEFGArsQQgQZCexCCBFkJLALIUSQkcAuhBBBRgK7EEIEGQnsQggRZCSwCyFEkJHALoQQQUYCuxBCBBkJ7EIIEWSmDOxKqSyl1AGl1EdKqTKl1D+4t8crpfYrpSrc/8fNf3KFEEJMJZAS+yjwba31CqAE+Dul1ArgceBtrfUS4G3330IIIa6yKQO71rpRa33S/bgXOIfrRtY7gGfduz0L7JynNAohhJiGad1oQym1CFgFHANStNaN7qeagBQ/xzwMPAxgs837vbOFEOK6F3DnqVIqEngJ+KbWusfzOa21BrSv47TWT2utV2utV1ut1lklVgghxNQCCuxKKTuuoP47rfUf3JublVJp7ufTgJb5SaIQQojpCGRUjAJ+BZzTWv+bx1OvAg+6Hz8IvDL3yRNCCDFdgTR6bwC+BJxVSp1yb/sfwI+BF5VSDwG1wGfnJYVCCCGmZcrArrV+F1B+nr5rbpMjhBBitmTmqRBCBBkJ7EIIEWRkYPk8UkrhGgnq/3mLxYLWmrGxsQVM2fxz9bkzaf6nYrF8XO7QWgf0WlO959M9byDXJdBzWiwWwsLCAFd+rly5MvNEziOLxTLrz+NsrkOgx9rtdhwOB/DJfj+vBgnsc2j58uWsWrWKVatWYbfbOXz4MC+99JLXPtnZ2ezcuZPc3FzS09OJiIigr6+P+vp6Ll68SHl5OYcOHWJkZGTK84WEhPCZz3yGTZs2AdDR0cFTTz1FU1PTvORvKoWFhaxevZpVq1ahtWbPnj3s27dvwn4hISEUFhaaE9aamppobm5maGgIAKvVSn5+Pk899ZR5zLPPPsv+/ftpa2vzei2r1UphYSElJSWsW7eOlpYWfvazn3H58mWv/ZRShIWFccMNN5jbKioq6O7uNoNYVFQU69ev59FHHzX3efLJJ3nnnXdwOp3mtqSkJAoLCykuLmbFihVcunSJH/zgB5O+N1FRUWzevJmvfvWr5rbvfve7fPjhhwwPD096LEBYWBjf+MY3GBgY4O2336asrGzKYwIRGRnJ7bffTmFhIdnZ2aSmphIWFkZfXx9tbW2cOHGCPXv20NzcPGUh5aabbqKkpIQ1a9bQ1NTEb37zGyorK6dMw5IlS1i5ciWrVq0iJiaGkydP8swzz0x6TH5+Pvfeey933303AENDQ3znO9+hvLx8em9AkJLAPkcsFguPPPIIubm5REVFMTQ0RH19vdfzd911F1/60pdIT08nLCyMkJAQLBYLTqeTzMxMioqKaGpq4ujRowEFdrvdzoYNG1i5ciUAIyMj/O53v6O1tdUrEC0Eq9XKY489RnJyspn/M2fO+Nw3Pj6eRx99lPDwcACOHDnCG2+8wUcffQS4gkRkZCQpKSlmyT8iIgJfE9wiIiJ44okniI2NJSoqiszMTPN1PVksFlJSUnj88Y+XNNq1axf79u2js7MTcM2Mjo6OJjU1FXCVAo0StkEpxaZNm3jggQeIjo4mLCyMlBSfk64nnD86Otpr34iICK/agT9Wq5WHHnqI2267jZCQECIiInA6nZw/f37KYyezfPlyHnjgAZYvX05MTAyhoaE4HA7zM5mVlcXixYsBeO655yb9TIaGhvLEE08QExNDVFQUKSkpxMTETJkGi8XCV77yFQoLC4mKigKgv79/yuNCQkKIi4sz38/BwUEiIiICyfZ1QQL7HNFaExkZSUJCAg6Hg9DQUBISEggLC2N4eJi8vDwefPBBli9fjsViMQMWuD7cdrudiIgIxsbGzJLrVBwOBwUFBYSEhKCUIjQ0lMzMTGpqaujr65uvrPo0NjZGVFQUiYmJ2Gw2QkJCSEhIIDIyckJaHA4HixYtIjIyEoDa2lrzsWF8wPMXALXWREdHk5iYiFIKq9VKQkICdXV1XiVh4/3Jy8szt6WkpHgtc6GU8rouxnk9t2mtcTgcxMbGEhsbi9aa+Ph4kpOTaWnxP0fPSNv4bYHYsmULmzdvJikpCbvdzq233srQ0BD9/f1ehYfpyMvL42tf+xpFRUVERkZOeH9tNhs2m43Q0FASExOn/AFyOp3ExsaSkJCAUoqkpCRiYmKw2+2T/iBorYmIiCAhIYHQ0FDGxsZISEggISGB9vZ2v8dZLJYJaQr0/bweSOfpHNFa09DQYAZlm81GTEwMycnJhIaGsnnzZm644QasVqv5Aezu7qahoYH6+no6OjoYHBykv7+f0dHRKc9ns9lITEwkNjbWa3t2dvaEILkQtNbU1taaabfZbGbAG28uv4BOp5Oamhrzb7vdTnp6us9S+1zp6uqitbUVcOXF4XCQn58/5XHj10oKtB07LS3NDOoAqampbNq0iS1btgRUKh7PYrGwfft2SkpKiIqKMgOk0+mkv7+ftrY2ent7p/Wa469DaGgoycnJU34WtdZcvnzZbB+3WCxERUWRnZ09ZR7GB/aFrqV+kkmJfQ7V1tayZs0as0ppfEDHxsbYsmWLud/IyAgXLlzg/PnzZttlTEwMMTExZrPAVIxS7/hS4KJFi65albS6upqbb76Z0NBQABITE8nMzOTixYte+wXS/DA++DudTp9tvE6nk6qqKtauXWtuW7x4Me+//z5dXV2TntfXD0wg5+3s7KSpqYklS5aYxxQXF3PkyBG/7dAWi8Xs6DOMjo4G1EnY19fntZ9SiuzsbLZu3UpjYyNvvfVWQE13xrFpaWncc8892Gw2M78tLS3U1tZSX19PT08P4eHhpKamEhsbS3Nz85Q/QlprKioqKC4uNl8zIyOD2NjYKT/T9fX19PX1ER8fD7i+N4WFhXzwwQd+j7FareYPnSGQAtH1QgL7HKqpqWFwcND8OzIykry8PEZHR72aAOrq6vjFL35BaWmpV3uiUirgFTBDQkLM9k9POTk5V6XEDlBVVeUVYFJTU8nPz+fw4cNegSkkJMTrOJvN5vUDpZSaUOIeGBjwWSIbGxub0EG3YsUKEhISqK+vN8+rlJpwXrvd7hXsjaYHT77ae43AbrBYLGzatIlnn33Wb0nXZrORkJDgtW14eHjKwB4WFkZmZuaEH3Cr1UpOTg4PPPAA9fX1lJWVBTwyZ926dV79Fz09PRw6dIjXXnuNsrIyRkdHUUoRExPD0qVL6enpmbI0rLWmvLzcKw3p6ekTapS+1NfXe73PMTExrF27lj/84Q9+R7pERESYBShDoE2Y1wNpiplD1dXVDAwMmB/u6Oho8vPzvTrjxsbG+O1vf8vx48cnBA2tdcAlr5CQELPE6CknJ4eoqKir0t5oBHYj/6mpqSxbtoy4OO+ba41vs42KiiI6Otr822q1TvjR6u3t9fneOJ1OKioqgI+HVi5btozs7Gyvjk+r1UpiYuKEdHiW+ozOV09dXV0TAmZHRweNjY3mdovFQkFBARs2bJjQ2WqcOz4+nuLiYnOb1nrKgGmxWFi7di3333//hB8ccNXaFi9ezKOPPhpQADVe07OQAXD48GF2797N6dOnzVKv1pquri6OHTvGuXPnAmo2MgK78b5kZmZOuPa+1NXVedVKIiIiWL58OSUlJRNK5eDKd25urvkZMb5XPT09E/a9Xklgn0M1NTX09/d7BfalS5d6BYv+/n7eeustBgYGZnWusLAwCgsLJ2yPjIwkIyPDK1AulIsXL3rVWBwOBzfeeCNf/OIXzWq/Ubr1rJksXryYgoICQkNDzSYLz2GJWmu6u7t9BvaxsTGqqqq8AmR4eDg7d+6kpKTE7NMIDw/nlltu8Tp2zZo1JCcnmyX3+Ph4r2ABriDuqynm8uXLE4LdE088wa233mp2Rhr/0tPT2bx5Mzk5OeZr9/f3097e7jewG00mP/jBD3A4HObYbqfT6dU8ZLzH3/rWtwL6MVdKmekwHD58eMadsAatNZWVlV7NIYEGdqP5x/N9TkhI4Hvf+x7r16/HarWa76XVauXmm29m7dq1JCUlmec2+gaEizTFzKHh4WEuXbpEfn4+UVFR2O12lixZ4lWybmtrm/Xkj4iICPLy8nx2TIKrKeLs2bN0d3fP6jzTNTQ0RF1dHUlJSWZTSmpqKp///OdZsWIFBw8eJD09nfvuu8+rxB4ZGcmnPvUpli1bRlVVlTlG3AhU/f39tLS0+K1qDw4OcvHiRfLz881jioqKSE5O5vbbb6eiooJFixaxY8cOr+Pi4+P5/ve/zwcffIDFYqGwsNCrptDX10dTU9OEwO50Ounq6uLSpUtmJ5/x4/HjH//Y7D8ZGBggLS2NpUuXmrU2w7vvvjtp00F0dDQ/+tGPiIuLM/M0ODjIT3/6U8LDw9m2bZtZ8rbb7Wzfvp0///nPHDhwYNLPl1JqQrDt6emZshkjKyuL0dFR2tvb/Y67Hxoa4uLFiyxduhSr1Up4eDjJycnExsZO6O/wNDw8TGNjI93d3WbajB/ap556ijNnzvDhhx9is9lYtmwZeXl5Xs2Ng4ODvPfee0E3yW82JLDPserqalauXOm3OSQrK4t7772Xffv20d3d7RU0HA4HaWlpxMTE+B0DDpjVen8ltMLCQjIyMsxx4QupsrKS5cuXm4HdmBi0bt061qxZY5baPRnBZv369ZSUlEwYdnjgwAE6Ojr8fnG11pw7d84rKCulSE9PJzU1Fa21347SnJwcsrKyzL+N/ZxOJ3v37vWqgXjq6emhurraa/SGcWxBQYFXx6qvYZR79uzxGyBjYmL4/Oc/zw033OB13C9+8Qveeecdent76ezs5Otf/7rZvKSUMic8tbS0+G1vN0Zvedb2Vq5cSU1NzYRJXcbrfu5zn+Phhx+mvr6eH/3oR5OOnzeug1FTSklJITk5edLADq5Se3t7u9ePjlFLufHGG80anK/3c2BggL179076+tebgAO7UsoKlAKXtNbblVK5wAtAAvA+8CWt9dRT6IKcUa30x2q18sgjj/DlL3+Z9vZ2BgYGzNJeVFQUISEhVFdX87d/+7d+XyMuLs6clKS1pq+vj1OnTrFu3TocDgfZ2dlkZGQQEhIy7Q4lY7x3WloaXV1ddHR0TOv4ixcvTujwMr6I4wP64OAgFosFm802Yby45z5vvPHGlLWPCxcusG3btgnnHd/pqLVmcHAQu91uBh9f+1y5coWXX37Z7/l6e3upra31+dxko36cTie7d+/m2LFjPkdxOBwO8vLy+OxnP+v1Oi+99BL79u2jtbWVsbExDh06REhICN/85jfNTuH4+Hi+8Y1v8JOf/MTvZ3BsbIzq6mqvbffddx9Wq5U33njDfM5isZCamsojjzzCxo0bzc7KDRs20N3dTWNjo6+X58KFC2zdutX8OyIiIqDO/MuXL/v8rPn6UfTU2trKH//4R0pLS6c8x/VkOiX2f8B1I2uj8fYnwL9rrV9QSv0n8BDw8zlO3zWnu7vbbzA1PqBRUVHmpAyjFGoEuLGxsUknZtjtdpKSksxquNaaU6dO8d5771FcXIzD4cBut5OdnU1WVlZAU7rBVUosKipi69atZGRkEBoayujoKJ2dnRw+fNjndH5fOjs7A+oAbm9v52c/+xkWi4Vt27ZRVFTkFWCN4Prkk09SVlY26bR7rTVtbW1TjgoZGxujra2Nf/7nfyY7O5uvfOUrxMbGegUOp9NJc3MzP//5zyd974aHh32WQj1H4XhyOp20tbWxZ88edu3a5bcmkJ6ezs6dO706QysqKti1axdNTU3m56Wzs5OjR4/yxz/+kc997nNmANy0aROvv/46H3zwgc9zjI2NcezYMf7mb/7GbLuPi4tj586drF+/nkuXLtHU1ITNZqOoqIisrCwiIyPNH+YtW7ZQUVHhN7C3t7dPGAE1fpinLz09PRP6ncYP8fQ0MDBAZWUl+/btY+/evQEPOrheBBTYlVKZwDbgh8Cj7rsq3Ql8wb3Ls8A/IYGdK1euTCiJaa0ZHh72Gm7na1wzuL54kw15TEhIIC8vz2sxqRMnTnDixAmGh4cJDw83xzkvWrQooMCenZ3NbbfdxpYtW8jNzTW/yOAKYKmpqeTk5PDKK69QXl4+6UiOK1eueD3f09PD+fPnOXXqlDkKpa6ujurqakpLS1FK0dDQQH5+PmlpaSQmJjI2Nsbly5epra3lvffemzCO25fxs1sbGho4c+YMra2tZkdyQ0MDFy9e5MSJE5w/f5729naysrJIS0sjPDycK1eu0NTURE1NDcePH5+0tjM6OuoViLTWnD17lgMHDpCQkEBMTAwhISGMjIzQ09NDe3s7NTU1lJeX+w2KMTEx3HTTTdxyyy3m4nBOp5Pnn3+euro6r8+V0+nk8uXL7N27l3Xr1rFo0SLA1Ta/detWamtrfTatGMND9+zZw/bt282hpomJicTFxZGTk0N/fz8Wi4WEhIQJC7F1dHRMutjW+GvlcDgmDDP1ZXBw0OvHe2hoiA8//JCDBw+SkZFBTEwMNpuN/v5+urq6aGpqora2lurq6kln/F6vAi2x/2/gvwPGwNEEoEtrbXzSGoAMXwcqpR4GHoaJM++C0eDgoNcXcHBwkOrqav785z+Tm5trrmUSHh5OWFgYVquV4eFh+vv76e3tpaenh6qqKr+vn5aWxrJly7xGSZSWltLQ0EBjYyORkZHY7XaysrLIzc3FarVOGogjIiJYu3Yt27dvp6CgYELJyGgaiI2Npa+vz8yPPwMDA15t4SMjI1y6dImXX36Z2NhYbDYbzc3NXp3IpaWlnD9/nvj4eGJjY82SdVtbW0ALZMHE8eYDAwOcPHmS48ePm00BLS0tZolyaGiIt99+m7i4OHM6++DgIB0dHQHVOsYHdnANjfzTn/5EeHi4eR1GR0fp7++np6eHjo6OSa9FVlYWJSUl5mgPgJMnT3Lo0CGfPzJDQ0NUVFTw+uuv89WvftVsWiopKeGtt96is7PT5+irvr4+XnrpJSIjI7n55puJjo7GarVitVqJjIz02XTidDo5fPgwe/fu9ZphOp7nqDBwfX4CKbEPDQ15vedOp5P29nZeffVVkpKSiIyMxGq1MjAwQF9fH11dXfT19UmHqR9TRlql1HagRWv9vlLq9umeQGv9NPA0QGho6OzWU70GjA/sw8PDlJeX88wzz5gjZKKjo4mKiiIyMhKbzcbg4CDd3d10dnbS2dnpt0Rns9nIyspi6dKlgKsE1d7eTlVVFUNDQ5SVlZGZmYndbichIYFFixaRnJzs9/XAtWbIunXrzBElxlj6lpYWQkJCzDVY4uPjue2226itraW2ttbvF2r8RCJjadXGxka/6TDGIM9mHPL4UqSxRk9dXZ3fY0ZHR2ltbTWXB5gOp9M5oakjNjaW3t5empubp/16drudvLw8ioqKgI+H8O3evXtC84an/v5+3nzzTbZs2UJBQQEAycnJFBcXU1NT4zf/Z8+e5fe//z2tra0UFBQQHx9vfiZ9jZn/6KOP2LVrF6dPn560xH7lypU5CexWq5Xo6Gi6u7sXfHRXMAj0nqf/TSl1DxCKq439KSBWKWVzl9ozgUvzl8xrx+DgoFdgM9YSGR4epqysbFbLrSYkJLB48WLS09MBV2A6c+aMWZp7//332bRpE9HR0dhsNrKzs7npppsmDewrV640RzEYTUZlZWX85S9/IS4ujrvvvpu4uDhzYsvixYsJDw/3u8jY+BK7zWab13VbDOMDirFS4XzxFdijo6NnPDEsKSmJxYsXk5aWZr5+eXk577zzzqTNUGNjYzQ3N/PKK6/wne98B3B95tasWcNf/vKXSX/YTp48SVlZGTfeeCNLliyhoKCAm266yWt0kTFR6bnnnuPUqVNTzr+YTYnds0BkrBkjZmbKCUpa6+9qrTO11ouA+4F3tNZfBA4An3bv9iDwyryl8hoyvsTueXOF2VqxYgXLli0z2z1HRkY4duyY+fypU6e82jgzMzO5+eabJx2lkZKS4jXErK2tjX/8x3/kl7/8Jf/6r//KgQMHzABmzKD0N34eJrax++tLmGvjf2iMUS/zxVdTzGzyuXTpUpYvX27WmoaHh3nttdcCWv9kZGSEN99806skXVBQQHp6+pTt20NDQ5SWlvL888/z8ssvew1lNJr6XnvtNQ4ePBjQpLqZBvbBwUGvErtRIBIzM5uZp4/h6kitxNXm/qu5SdK1bWBgwGtavdVq9Vm1nS5jAs3y5cuBj5cfOHLkiLmP0fFnfMHj4uIoKCgwx2n74rlOizEKxrPTraamxiu4eN61xhejxuKZ/4X4gholds8ZmfPZpzM6Osrg4KBXEPNcVGu6li1bZl7bsbExOjs7J4zN9hzDPf48nZ2dXpOT7Hb7hFnPkwkPD6ekpIQ77rgD+PiOVefPn+c//uM/Ah42O/46hISEBNR5ajTFjH8/xcxMK7BrrQ9qrbe7H1/UWq/VWudrrT+jtZYVeHB94UdGRsxSq6/Fp2bCmPRiVE+HhoY4d+7chGaWY8eOebUZJycnc9ddd/l9Xc/RCFarlbi4OFasWAG4vuz33HOPV42ju7t70uGYIyMjDA8PmwHG1/j1+TAyMjKhxDef6+UYpWrPc/pa1yQQYWFhxMfHm52W/f39HDx4cEJb9sqVK9m4cSMbN25kxYoVXqt4aq154403vGpL42tjk7njjjvYuHGj17UeHh7mhz/8YcAd2OD6UfJcLykkJCSggo3T6WR4eNjreyOBfebknZsHRrXSmHgzF00xGzZsIDc312st99dff33Cfu+88w533HEH2dnZWCwWkpKSuP3223nuued8fkHr6upoaWkxO0nT09P59a9/TVVVFenp6V7txgMDA7S2tk45nt1ojjJqAgu1IFlfX5853HO+Azu4gtiVK1fMGslMA1FiYqLXTOXh4WFzYTNDbGwsjz32mDmjtbS0lF/96lecOHHC3KeiosKrf8O4q9RUUlJSuO222ygsLPS61uObZgLV39/P2NgYFosl4HHs4Mr30NCQWfORwD5zsgjYPBgaGjKDqPHhno2UlBRWr17t1Wna0tLCoUOHJuzb0dHB6dOnzeYUm81Gamqq13rwng4fPmxOZjGCoXFnJiOoG1/2AwcOUFpaOuWYcqM5Cua/5OzJs313Ic5pBHbDTNcDT0xM9BpiaIzW8bRkyRIiIiLMOxvFxMRMuCWfMZzSeA+ioqIC6ri+9957zTt7gav209DQwK9//esZ3ZDaCOwQ+Dh2+Diww8ft+2JmJLDPg5GREfNLPhdNEZs3b/Zak7utrY2jR4/6HJmiteb48ePmWHNjXe0dO3b47Ezs6Ohg//797N+/3+wkNUpLntP8T548yZtvvsmFCxemTK9nldp4vYXg2ea9UCV2z5ExM+1LGd8fYLfbJ7SNj79bUnp6OsXFxV6frfT0dK9rPDo6OuWPTV5eHuvXryc1NdV8v9ra2njhhRemvZyEYWhoyGtJ40A//57NaXPVhHm9krrOPBgbG/OqEs8mwMTExLBp0yazqURrTUtLC++++67f0lRFRQWVlZUUFRWZq0zm5+dTXFzsVXU30lpRUcGLL75IfX09GzduJCMjg/j4eAYHB2lsbKS0tJRDhw5x9uxZv1Phx7/mTEp6s7XQk1WMdcAN/ta7mcqVK1e8msnCw8PZuHEjVVVVNDQ0UFJSQklJiVebemRkJEVFRXzmM59h//79hIWFsWPHDq8O3N7e3ilHsmzdupXs7Gyzf6Cvr49z585x6NChGb+fnp2n03k/PI+DwO60JXyTwP4Jt3btWnJzc83SYG9vL9XV1ZMuFdDX10d5eTk1NTXceOONWCwWIiMj2bZtm8+mlP7+fs6dO0dbWxsXL14076U6ODhIe3s7Fy5coLa2Vu5QM0+6urq8mnSMNdb/+q//mra2NpYvX056erpXqd5qtZKens5f/dVfsWjRIhwOx4ShrZ2dnZPeuzQnJ4dbbrnFqybQ2NjIwYMHr9ra5lejQBCMJLB/goWEhHD33Xd7daxdvnyZM2fO+Lxlm6cLFy5QVlbGihUrsFqt2Gw2brnlFjIyMmhoaJiw/+joqNfsUIvFItO1F0hnZyctLS309fWZU+djY2O5/fbbJ+zb0dGB3W43Z4guXbrUnInsyZh1629WrVKKu+66i6ysLLNzs6+vj/Pnz3P06NE5zZ9YeFLX+QRLSUmhuLjYHFUzOjpKdXX1pDf5NdTX13P+/HlzOrbFYiExMZH169cHVMWVoL5wjOaPysrKSTsMe3t7OXLkCGfOnPE78xdc166uro4zZ874nXUcFxfHtm3bzM5VrTXV1dUcP35cFtUKAlJinwdGG7vnv5nIzs5mYGCAsLAwHA4HnZ2dVFRUTLoIl2FkZMT8EbjzzjtxOp309/eTlZW1IJ2Kc5H/6XI6nV7nXIhqvXFO4/FMz3ns2DESEhJISkoiPj7evF2f1prR0VGGh4c5cuQIv/3tb4mJieG+++5jw4YN2O127HY7SimcTiejo6N0dXWxe/duzp8/7/eHYv369aSlpZmdrUNDQ+byz7NlvPfTvQ7jr52Mipk5CezzoLu7m9bWViwWC6Ojo153tJ+Od999l29/+9t84QtfoKSkhFOnTnH27NmAvyg1NTW89957rF+/nrq6On7zm9+wb9++eQ+0HR0dtLa24nQ6GRkZmdEiWzPR3Nxs3sPUWC1zPo2MjNDc3Ex8fDzAhDtiTUdHRwcvvvgip0+fZufOnSxdupS0tDT6+/spLy/n6NGjHD582Fx3vrm5mQ8++IBVq1axatUqwsLCaGhooLKykt///vdUVVVNOrGopaWFmpoacnNzcTgcnDp1ihMnTsx4JMz4125rayM0NJShoaEp755k6O3tpaWlBYfDwdjY2Iy/NwLUQnZWhIaG6vE30hWBycrKYmRkhI6OjmnNBIyKiiIrK4sLFy5ICUh4UUqxYcMG7rzzTg4fPszRo0dnfZN1MT/Ky8vf11qvDnR/CexCXMeM8f7jhxqKT5bpBnZpihHiOiYBPTjJqBghhAgyAQV2pVSsUmq3Uuq8UuqcUmq9UipeKbVfKVXh/j+wZeSEEELMq0BL7E8Bb2qtlwErgXPA48DbWuslwNvuv4UQQlxlUwZ2pVQMcCvuG2lorYe11l3ADuBZ927PAjvnJ4lCCCGmI5ASey7QCvxaKfWBUuqXSqkIIEVrbUxrawJSfB2slHpYKVWqlCqV4XZCCDH/AgnsNqAY+LnWehXQz7hmF+3qVvfZta61flprvVprvXo+70EphBDCJZDA3gA0aK2NuybvxhXom5VSaQDu/2WBCSGE+ASYMrBrrZuAeqWUsYTcXcBHwKvAg+5tDwKvzEsKhRBCTEugE5QeAX6nlHIAF4Ev4/pReFEp9RBQC3x2fpIohBBiOgIK7FrrU4Cv6ax3zWlqhBBCzJrMPBVCiCAjgV0IIYKMBHYhhAgyEtiFECLISGAXQoggI4FdCCGCjAR2IYQIMhLYhRAiyEhgF0KIICOBXQghgowEdiGECDIS2IUQIshIYBdCiCATUGBXSn1LKVWmlPpQKfW8UipUKZWrlDqmlKpUSu1yL+krhBDiKgvkZtYZwDeA1VrrGwArcD/wE+Dftdb5QCfw0HwmVAghRGACbYqxAWFKKRsQDjQCd+K6TR7As8DOOU+dEEKIaQvk1niXgP8F1OEK6N3A+0CX1nrUvVsDkOHreKXUw0qpUqVUqdPpnJtUCyGE8CuQppg4YAeQC6QDEcDdgZ5Aa/201nq11nq11WqdcUKFEEIEJpCmmM1Atda6VWs9AvwB2ADEuptmADKBS/OURiGEENMQSGCvA0qUUuFKKYXrPqcfAQeAT7v3eRB4ZX6SKIQQYjoCaWM/hquT9CRw1n3M08BjwKNKqUogAfjVPKZTCCFEgJTWesFOFhoaqnNychbsfEIIEQzKy8vf11qvDnR/mXkqhBBBRgK7EEIEGQnsQggRZCSwCyFEkJHALoQQQUYCuxBCBBkJ7EIIEWQksAshRJCRwC6EEEFGArsQQgQZCexCCBFkJLALIUSQkcAuhBBBRgK7EEIEGQnsQggRZCSwCyFEkFnQG20opXqBCwt2woWVCLRd7UTMo2DOn+Tt2nQ95S1Ha50U6MG2qXeZUxemcxeQa4lSqjRY8wbBnT/J27VJ8uafNMUIIUSQkcAuhBBBZqED+9MLfL6FFMx5g+DOn+Tt2iR582NBO0+FEELMP2mKEUKIICOBXQghgsyCBXal1N1KqQtKqUql1OMLdd75opSqUUqdVUqdUkqVurfFK6X2K6Uq3P/HXe10BkIp9YxSqkUp9aHHNp95US4/dV/HM0qp4quX8qn5yds/KaUuua/dKaXUPR7PfdedtwtKqf96dVIdGKVUllLqgFLqI6VUmVLqH9zbr/lrN0nervlrp5QKVUodV0qdduftf7q35yqljrnzsEsp5XBvD3H/Xel+ftGUJ9Faz/s/wApUAXmAAzgNrFiIc89jnmqAxHHb/gV43P34ceAnVzudAeblVqAY+HCqvAD3AHsABZQAx652+meQt38CvuNj3xXuz2YIkOv+zFqvdh4myVsaUOx+HAWUu/NwzV+7SfJ2zV879/sf6X5sB465r8eLwP3u7f8JfM39+OvAf7of3w/smuocC1ViXwtUaq0vaq2HgReAHQt07oW0A3jW/fhZYOfVS0rgtNaHgI5xm/3lZQfwW+1yFIhVSqUtSEJnwE/e/NkBvKC1HtJaVwOVuD67n0ha60at9Un3417gHJBBEFy7SfLmzzVz7dzvf5/7T7v7nwbuBHa7t4+/bsb13A3cpZRSk51joQJ7BlDv8XcDk1+ka4EG9iml3ldKPezelqK1bnQ/bgJSrk7S5oS/vATLtfx7d3PEMx5NZtds3tzV81W4Sn9Bde3G5Q2C4NoppaxKqVNAC7AfVw2jS2s96t7FM/1m3tzPdwMJk72+dJ7O3EatdTGwFfg7pdStnk9qV70pKMaSBlNe3H4OLAaKgEbgyauamllSSkUCLwHf1Fr3eD53rV87H3kLimuntXZqrYuATFw1i2Vz+foLFdgvAVkef2e6t12ztNaX3P+3AC/jujjNRtXW/X/L1UvhrPnLyzV/LbXWze4v1hjwCz6usl9zeVNK2XEFvt9prf/g3hwU185X3oLp2gForbuAA8B6XE1jxvpdnuk38+Z+PgZon+x1FyqwnwCWuHt9Hbg6AF5doHPPOaVUhFIqyngM/BfgQ1x5etC924PAK1cnhXPCX15eBR5wj7AoAbo9qv3XhHHtyp/Cde3Albf73aMQcoElwPGFTl+g3O2svwLOaa3/zeOpa/7a+ctbMFw7pVSSUirW/TgM2IKrD+EA8Gn3buOvm3E9Pw28466J+beAPcH34OrZrgK+t9A90XOclzxcPfCngTIjP7javd4GKoC3gPirndYA8/M8rmrtCK62vYf85QVXj/7/cV/Hs8Dqq53+GeTt/7rTfsb9pUnz2P977rxdALZe7fRPkbeNuJpZzgCn3P/uCYZrN0nervlrB9wEfODOw4fA993b83D9GFUC/w8IcW8Pdf9d6X4+b6pzyJICQggRZKTzVAghgowEdiGECDIS2IUQIshIYBdCiCAjgV0IIYKMBHYhhAgyEtiFECLI/H/jYdb6S3qenQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "plt.imshow(numbers_image)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "fourth-narrow",
   "metadata": {},
   "outputs": [],
   "source": [
    "letters_image = cv2.imread('letter_inference_2.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "uniform-video",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fc59563e400>"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "plt.imshow(letters_image)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "wrong-halloween",
   "metadata": {},
   "source": [
    "## 3.6. Run the inference"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "every-spell",
   "metadata": {},
   "outputs": [],
   "source": [
    "%time\n",
    "\n",
    "numbers_result = number_model(numbers_image)\n",
    "letter_result = letter_model(letters_image)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "acquired-validation",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'۴۸۰۳۳۲۵۶۱۲'"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "extract_chars(numbers_result)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "impaired-deficit",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'متاثر'"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "extract_letters(letter_result)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "imported-short",
   "metadata": {},
   "source": [
    "# Final Thoughts"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "korean-progressive",
   "metadata": {},
   "source": [
    "This notebook is just a tool to put what we've done to the test. It's not official release of the tool yet and we're going to work on some other tools including:\n",
    "\n",
    "1. A CLI tool (similar to _tesseract_)\n",
    "2. A web service\n",
    "3. A python library for developers\n",
    "\n",
    "And of course there'll be changes in pretty much everything you've seen in the repo. We're trying to improve the product for good."
   ]
  }
 ],
 "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": "3.8.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}