deeplab+fake_detection.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "initial_id",
   "metadata": {
    "collapsed": true,
    "ExecuteTime": {
     "end_time": "2024-04-23T09:29:48.173547Z",
     "start_time": "2024-04-23T09:29:46.045369Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-04-23 11:29:46.269521: E tensorflow/compiler/xla/stream_executor/cuda/cuda_dnn.cc:9342] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered\n",
      "2024-04-23 11:29:46.269548: E tensorflow/compiler/xla/stream_executor/cuda/cuda_fft.cc:609] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered\n",
      "2024-04-23 11:29:46.269575: E tensorflow/compiler/xla/stream_executor/cuda/cuda_blas.cc:1518] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered\n",
      "2024-04-23 11:29:46.276190: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.\n",
      "To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "\n",
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "\n",
    "from tensorflow import keras\n",
    "from tensorflow.keras import layers\n",
    "from deeplab_mdl_def import DynamicUpsample\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator"
   ]
  },
  {
   "cell_type": "code",
   "outputs": [],
   "source": [
    "def DeeplabV3Plus_nicco(num_classes,\n",
    "                        filters_conv1=24, filters_conv2=24,\n",
    "                        filters_spp=128, filters_final=128,\n",
    "                        dilated_conv_rates =[1, 4, 8, 16],\n",
    "                        trainable_resnet=False,\n",
    "                        path_model_trained: str = None,\n",
    "                        transfer_learning: str = \"last\"):\n",
    "    \n",
    "    assert transfer_learning in [\"last\", \"all\"], \"NICCO: Transfer learning policy not supported\"\n",
    "    model_input = keras.Input(shape=(None, None, 3))\n",
    "    preprocessed = keras.applications.resnet50.preprocess_input(model_input)\n",
    "    resnet50 = keras.applications.ResNet50(weights=\"imagenet\", include_top=False, input_tensor=preprocessed)\n",
    "    resnet50.trainable=trainable_resnet\n",
    "\n",
    "    x = resnet50.get_layer(\"conv4_block6_2_relu\").output\n",
    "    input_b = resnet50.get_layer(\"conv2_block3_2_relu\").output\n",
    "\n",
    "    x1 = layers.GlobalAveragePooling2D()(x)\n",
    "    x1 = layers.Reshape((1, 1, x.shape[-1]))(x1)\n",
    "    x1 = layers.Conv2D(filters=filters_conv1, kernel_size=1, padding=\"same\")(x1) #Perché qui c'è una conv2d?\n",
    "    x1 = layers.BatchNormalization()(x1)\n",
    "    x1 = DynamicUpsample()(x1, x)\n",
    "\n",
    "    # Multiple dilated convolutions\n",
    "    pyramids = []\n",
    "    for rate in dilated_conv_rates:\n",
    "        if rate == 1:\n",
    "            pyramid = layers.Conv2D(filters=filters_spp, kernel_size=3, dilation_rate=rate, padding=\"same\")(x)\n",
    "            pyramid = layers.BatchNormalization()(pyramid)\n",
    "            pyramids.append(pyramid)\n",
    "        else:\n",
    "            pyramid = layers.Conv2D(filters=filters_spp, kernel_size=3 + int(rate*(1/3)), dilation_rate=rate, padding=\"same\")(x)\n",
    "            pyramid = layers.BatchNormalization()(pyramid)\n",
    "            pyramids.append(pyramid)\n",
    "\n",
    "    x = layers.Concatenate(axis=-1)([x1] + pyramids)\n",
    "    x = layers.Conv2D(filters=filters_spp, kernel_size=1, padding=\"same\")(x)\n",
    "    x = layers.BatchNormalization()(x)\n",
    "\n",
    "    input_b = layers.Conv2D(filters=filters_conv2, kernel_size=1, padding=\"same\")(input_b)\n",
    "    input_b = layers.BatchNormalization()(input_b)\n",
    "\n",
    "    input_a = DynamicUpsample()(x, input_b)\n",
    "\n",
    "    x = layers.Concatenate(axis=-1)([input_a, input_b])\n",
    "    x = layers.Conv2D(filters=filters_final, kernel_size=3, padding=\"same\")(x)\n",
    "    x = layers.BatchNormalization()(x)\n",
    "    x = layers.Conv2D(filters=filters_final, kernel_size=3, padding=\"same\")(x)\n",
    "    x = layers.BatchNormalization()(x)\n",
    "    x = DynamicUpsample()(x, model_input)\n",
    "    \n",
    "    #resnet_output = resnet50.get_layer(\"conv4_block6_2_relu\").output\n",
    "    #resnet_output = layers.GlobalAveragePooling2D()(resnet_output)\n",
    "    x = layers.GlobalAveragePooling2D()(x)\n",
    "    #x = layers.Concatenate(axis=-1)([x, resnet_output])\n",
    "    x = layers.Dense(50,activation=\"selu\")(x)\n",
    "    x = layers.Dense(25,activation=\"selu\")(x)\n",
    "    model_output = layers.Dense(num_classes,activation=\"sigmoid\")(x)\n",
    "    model = keras.Model(inputs=model_input, outputs=model_output)\n",
    "    \n",
    "    if path_model_trained is not None:\n",
    "        model_from = tf.keras.models.load_model(path_model_trained, \n",
    "                                                custom_objects={'DynamicUpsample': DynamicUpsample})\n",
    "        for l, l_old in zip(model.layers[:-3], model_from.layers):\n",
    "            l.set_weights(l_old.get_weights())\n",
    "            if transfer_learning==\"last\": l.trainable = False\n",
    "    return model\n"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-04-23T09:52:14.587363Z",
     "start_time": "2024-04-23T09:52:14.570858Z"
    }
   },
   "id": "c3f3199b009d3b8c",
   "execution_count": 22
  },
  {
   "cell_type": "code",
   "outputs": [],
   "source": [
    "model_name : str = \"deeplabv3plus_face_segmentation_augmentation_class_weights_latest_fixConv_different_rate.h5\""
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-04-23T09:52:15.410544Z",
     "start_time": "2024-04-23T09:52:15.405903Z"
    }
   },
   "id": "c69b61a6fcc75a74",
   "execution_count": 23
  },
  {
   "cell_type": "code",
   "outputs": [],
   "source": [
    "model = tf.keras.models.load_model(model_name, custom_objects={'DynamicUpsample': DynamicUpsample})"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-04-23T09:52:18.114927Z",
     "start_time": "2024-04-23T09:52:15.529606Z"
    }
   },
   "id": "31ecfd9badbbc296",
   "execution_count": 24
  },
  {
   "cell_type": "code",
   "outputs": [],
   "source": [
    "model_nicco = DeeplabV3Plus_nicco(1)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-04-23T09:52:19.770337Z",
     "start_time": "2024-04-23T09:52:18.117578Z"
    }
   },
   "id": "ada74431800b27a3",
   "execution_count": 25
  },
  {
   "cell_type": "code",
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(7, 7, 3, 64) | (7, 7, 3, 64) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(1, 1, 64, 64) | (1, 1, 64, 64) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(3, 3, 64, 64) | (3, 3, 64, 64) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(1, 1, 64, 256) | (1, 1, 64, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 64, 256) | (1, 1, 64, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 256, 64) | (1, 1, 256, 64) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(3, 3, 64, 64) | (3, 3, 64, 64) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(1, 1, 64, 256) | (1, 1, 64, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 256, 64) | (1, 1, 256, 64) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(3, 3, 64, 64) | (3, 3, 64, 64) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(64,) | (64,) | True | False\n",
      "(1, 1, 64, 256) | (1, 1, 64, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 256, 128) | (1, 1, 256, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(3, 3, 128, 128) | (3, 3, 128, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(1, 1, 256, 512) | (1, 1, 256, 512) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(1, 1, 128, 512) | (1, 1, 128, 512) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(1, 1, 512, 128) | (1, 1, 512, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(3, 3, 128, 128) | (3, 3, 128, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(1, 1, 128, 512) | (1, 1, 128, 512) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(1, 1, 512, 128) | (1, 1, 512, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(3, 3, 128, 128) | (3, 3, 128, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(1, 1, 128, 512) | (1, 1, 128, 512) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(1, 1, 512, 128) | (1, 1, 512, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(3, 3, 128, 128) | (3, 3, 128, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(1, 1, 128, 512) | (1, 1, 128, 512) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(512,) | (512,) | True | False\n",
      "(1, 1, 512, 256) | (1, 1, 512, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(3, 3, 256, 256) | (3, 3, 256, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 512, 1024) | (1, 1, 512, 1024) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1, 1, 256, 1024) | (1, 1, 256, 1024) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1, 1, 1024, 256) | (1, 1, 1024, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(3, 3, 256, 256) | (3, 3, 256, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 256, 1024) | (1, 1, 256, 1024) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1, 1, 1024, 256) | (1, 1, 1024, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(3, 3, 256, 256) | (3, 3, 256, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 256, 1024) | (1, 1, 256, 1024) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1, 1, 1024, 256) | (1, 1, 1024, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(3, 3, 256, 256) | (3, 3, 256, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 256, 1024) | (1, 1, 256, 1024) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1, 1, 1024, 256) | (1, 1, 1024, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(3, 3, 256, 256) | (3, 3, 256, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 256, 1024) | (1, 1, 256, 1024) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1024,) | (1024,) | True | False\n",
      "(1, 1, 1024, 256) | (1, 1, 1024, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(3, 3, 256, 256) | (3, 3, 256, 256) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(256,) | (256,) | True | False\n",
      "(1, 1, 256, 24) | (1, 1, 256, 24) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(3, 3, 256, 128) | (3, 3, 256, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(4, 4, 256, 128) | (4, 4, 256, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(5, 5, 256, 128) | (5, 5, 256, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(8, 8, 256, 128) | (8, 8, 256, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(1, 1, 536, 128) | (1, 1, 536, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(1, 1, 64, 24) | (1, 1, 64, 24) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(24,) | (24,) | True | False\n",
      "(3, 3, 152, 128) | (3, 3, 152, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(3, 3, 128, 128) | (3, 3, 128, 128) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(128,) | (128,) | True | False\n",
      "(384, 100) | (1, 1, 128, 11) | False | False\n",
      "(100,) | (11,) | False | False\n"
     ]
    }
   ],
   "source": [
    "for i, j in zip(model_nicco.get_weights(), model.get_weights()): \n",
    "    print(i.shape, \"|\", j.shape,  \"|\", np.array_equal(i.shape, j.shape), \"|\", np.array_equal(i,j))"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-04-23T09:32:01.594514Z",
     "start_time": "2024-04-23T09:32:01.362341Z"
    }
   },
   "id": "ce9e4305b7986f69",
   "execution_count": 10
  },
  {
   "cell_type": "code",
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "input_2 False []\n",
      "tf.__operators__.getitem_2 False []\n",
      "tf.nn.bias_add_2 False []\n",
      "conv1_pad False []\n",
      "conv1_conv False [(7, 7, 3, 64), (64,)]\n",
      "conv1_bn False [(64,), (64,), (64,), (64,)]\n",
      "conv1_relu False []\n",
      "pool1_pad False []\n",
      "pool1_pool False []\n",
      "conv2_block1_1_conv False [(1, 1, 64, 64), (64,)]\n",
      "conv2_block1_1_bn False [(64,), (64,), (64,), (64,)]\n",
      "conv2_block1_1_relu False []\n",
      "conv2_block1_2_conv False [(3, 3, 64, 64), (64,)]\n",
      "conv2_block1_2_bn False [(64,), (64,), (64,), (64,)]\n",
      "conv2_block1_2_relu False []\n",
      "conv2_block1_0_conv False [(1, 1, 64, 256), (256,)]\n",
      "conv2_block1_3_conv False [(1, 1, 64, 256), (256,)]\n",
      "conv2_block1_0_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv2_block1_3_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv2_block1_add False []\n",
      "conv2_block1_out False []\n",
      "conv2_block2_1_conv False [(1, 1, 256, 64), (64,)]\n",
      "conv2_block2_1_bn False [(64,), (64,), (64,), (64,)]\n",
      "conv2_block2_1_relu False []\n",
      "conv2_block2_2_conv False [(3, 3, 64, 64), (64,)]\n",
      "conv2_block2_2_bn False [(64,), (64,), (64,), (64,)]\n",
      "conv2_block2_2_relu False []\n",
      "conv2_block2_3_conv False [(1, 1, 64, 256), (256,)]\n",
      "conv2_block2_3_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv2_block2_add False []\n",
      "conv2_block2_out False []\n",
      "conv2_block3_1_conv False [(1, 1, 256, 64), (64,)]\n",
      "conv2_block3_1_bn False [(64,), (64,), (64,), (64,)]\n",
      "conv2_block3_1_relu False []\n",
      "conv2_block3_2_conv False [(3, 3, 64, 64), (64,)]\n",
      "conv2_block3_2_bn False [(64,), (64,), (64,), (64,)]\n",
      "conv2_block3_2_relu False []\n",
      "conv2_block3_3_conv False [(1, 1, 64, 256), (256,)]\n",
      "conv2_block3_3_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv2_block3_add False []\n",
      "conv2_block3_out False []\n",
      "conv3_block1_1_conv False [(1, 1, 256, 128), (128,)]\n",
      "conv3_block1_1_bn False [(128,), (128,), (128,), (128,)]\n",
      "conv3_block1_1_relu False []\n",
      "conv3_block1_2_conv False [(3, 3, 128, 128), (128,)]\n",
      "conv3_block1_2_bn False [(128,), (128,), (128,), (128,)]\n",
      "conv3_block1_2_relu False []\n",
      "conv3_block1_0_conv False [(1, 1, 256, 512), (512,)]\n",
      "conv3_block1_3_conv False [(1, 1, 128, 512), (512,)]\n",
      "conv3_block1_0_bn False [(512,), (512,), (512,), (512,)]\n",
      "conv3_block1_3_bn False [(512,), (512,), (512,), (512,)]\n",
      "conv3_block1_add False []\n",
      "conv3_block1_out False []\n",
      "conv3_block2_1_conv False [(1, 1, 512, 128), (128,)]\n",
      "conv3_block2_1_bn False [(128,), (128,), (128,), (128,)]\n",
      "conv3_block2_1_relu False []\n",
      "conv3_block2_2_conv False [(3, 3, 128, 128), (128,)]\n",
      "conv3_block2_2_bn False [(128,), (128,), (128,), (128,)]\n",
      "conv3_block2_2_relu False []\n",
      "conv3_block2_3_conv False [(1, 1, 128, 512), (512,)]\n",
      "conv3_block2_3_bn False [(512,), (512,), (512,), (512,)]\n",
      "conv3_block2_add False []\n",
      "conv3_block2_out False []\n",
      "conv3_block3_1_conv False [(1, 1, 512, 128), (128,)]\n",
      "conv3_block3_1_bn False [(128,), (128,), (128,), (128,)]\n",
      "conv3_block3_1_relu False []\n",
      "conv3_block3_2_conv False [(3, 3, 128, 128), (128,)]\n",
      "conv3_block3_2_bn False [(128,), (128,), (128,), (128,)]\n",
      "conv3_block3_2_relu False []\n",
      "conv3_block3_3_conv False [(1, 1, 128, 512), (512,)]\n",
      "conv3_block3_3_bn False [(512,), (512,), (512,), (512,)]\n",
      "conv3_block3_add False []\n",
      "conv3_block3_out False []\n",
      "conv3_block4_1_conv False [(1, 1, 512, 128), (128,)]\n",
      "conv3_block4_1_bn False [(128,), (128,), (128,), (128,)]\n",
      "conv3_block4_1_relu False []\n",
      "conv3_block4_2_conv False [(3, 3, 128, 128), (128,)]\n",
      "conv3_block4_2_bn False [(128,), (128,), (128,), (128,)]\n",
      "conv3_block4_2_relu False []\n",
      "conv3_block4_3_conv False [(1, 1, 128, 512), (512,)]\n",
      "conv3_block4_3_bn False [(512,), (512,), (512,), (512,)]\n",
      "conv3_block4_add False []\n",
      "conv3_block4_out False []\n",
      "conv4_block1_1_conv False [(1, 1, 512, 256), (256,)]\n",
      "conv4_block1_1_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block1_1_relu False []\n",
      "conv4_block1_2_conv False [(3, 3, 256, 256), (256,)]\n",
      "conv4_block1_2_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block1_2_relu False []\n",
      "conv4_block1_0_conv False [(1, 1, 512, 1024), (1024,)]\n",
      "conv4_block1_3_conv False [(1, 1, 256, 1024), (1024,)]\n",
      "conv4_block1_0_bn False [(1024,), (1024,), (1024,), (1024,)]\n",
      "conv4_block1_3_bn False [(1024,), (1024,), (1024,), (1024,)]\n",
      "conv4_block1_add False []\n",
      "conv4_block1_out False []\n",
      "conv4_block2_1_conv False [(1, 1, 1024, 256), (256,)]\n",
      "conv4_block2_1_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block2_1_relu False []\n",
      "conv4_block2_2_conv False [(3, 3, 256, 256), (256,)]\n",
      "conv4_block2_2_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block2_2_relu False []\n",
      "conv4_block2_3_conv False [(1, 1, 256, 1024), (1024,)]\n",
      "conv4_block2_3_bn False [(1024,), (1024,), (1024,), (1024,)]\n",
      "conv4_block2_add False []\n",
      "conv4_block2_out False []\n",
      "conv4_block3_1_conv False [(1, 1, 1024, 256), (256,)]\n",
      "conv4_block3_1_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block3_1_relu False []\n",
      "conv4_block3_2_conv False [(3, 3, 256, 256), (256,)]\n",
      "conv4_block3_2_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block3_2_relu False []\n",
      "conv4_block3_3_conv False [(1, 1, 256, 1024), (1024,)]\n",
      "conv4_block3_3_bn False [(1024,), (1024,), (1024,), (1024,)]\n",
      "conv4_block3_add False []\n",
      "conv4_block3_out False []\n",
      "conv4_block4_1_conv False [(1, 1, 1024, 256), (256,)]\n",
      "conv4_block4_1_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block4_1_relu False []\n",
      "conv4_block4_2_conv False [(3, 3, 256, 256), (256,)]\n",
      "conv4_block4_2_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block4_2_relu False []\n",
      "conv4_block4_3_conv False [(1, 1, 256, 1024), (1024,)]\n",
      "conv4_block4_3_bn False [(1024,), (1024,), (1024,), (1024,)]\n",
      "conv4_block4_add False []\n",
      "conv4_block4_out False []\n",
      "conv4_block5_1_conv False [(1, 1, 1024, 256), (256,)]\n",
      "conv4_block5_1_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block5_1_relu False []\n",
      "conv4_block5_2_conv False [(3, 3, 256, 256), (256,)]\n",
      "conv4_block5_2_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block5_2_relu False []\n",
      "conv4_block5_3_conv False [(1, 1, 256, 1024), (1024,)]\n",
      "conv4_block5_3_bn False [(1024,), (1024,), (1024,), (1024,)]\n",
      "conv4_block5_add False []\n",
      "conv4_block5_out False []\n",
      "conv4_block6_1_conv False [(1, 1, 1024, 256), (256,)]\n",
      "conv4_block6_1_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block6_1_relu False []\n",
      "conv4_block6_2_conv False [(3, 3, 256, 256), (256,)]\n",
      "conv4_block6_2_bn False [(256,), (256,), (256,), (256,)]\n",
      "conv4_block6_2_relu False []\n",
      "global_average_pooling2d_1 True []\n",
      "reshape_1 True []\n",
      "conv2d_9 True [(1, 1, 256, 24), (24,)]\n",
      "batch_normalization_9 True [(24,), (24,), (24,), (24,)]\n",
      "conv2d_10 True [(3, 3, 256, 128), (128,)]\n",
      "conv2d_11 True [(4, 4, 256, 128), (128,)]\n",
      "conv2d_12 True [(5, 5, 256, 128), (128,)]\n",
      "conv2d_13 True [(8, 8, 256, 128), (128,)]\n",
      "dynamic_upsample_3 True []\n",
      "batch_normalization_10 True [(128,), (128,), (128,), (128,)]\n",
      "batch_normalization_11 True [(128,), (128,), (128,), (128,)]\n",
      "batch_normalization_12 True [(128,), (128,), (128,), (128,)]\n",
      "batch_normalization_13 True [(128,), (128,), (128,), (128,)]\n",
      "concatenate_2 True []\n",
      "conv2d_14 True [(1, 1, 536, 128), (128,)]\n",
      "conv2d_15 True [(1, 1, 64, 24), (24,)]\n",
      "batch_normalization_14 True [(128,), (128,), (128,), (128,)]\n",
      "batch_normalization_15 True [(24,), (24,), (24,), (24,)]\n",
      "dynamic_upsample_4 True []\n",
      "concatenate_3 True []\n",
      "conv2d_16 True [(3, 3, 152, 128), (128,)]\n",
      "batch_normalization_16 True [(128,), (128,), (128,), (128,)]\n",
      "conv2d_17 True [(3, 3, 128, 128), (128,)]\n",
      "batch_normalization_17 True [(128,), (128,), (128,), (128,)]\n",
      "dynamic_upsample_5 True []\n",
      "global_average_pooling2d_3 True []\n",
      "global_average_pooling2d_2 True []\n",
      "concatenate_4 True []\n",
      "dense True [(384, 100), (100,)]\n",
      "dense_1 True [(100, 1), (1,)]\n"
     ]
    }
   ],
   "source": [
    "for l in model_nicco.layers:\n",
    "    print(l.name, l.trainable, [i.shape for i in l.get_weights()])"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-04-23T09:32:01.675519Z",
     "start_time": "2024-04-23T09:32:01.596165Z"
    }
   },
   "id": "d031c6eca8892bef",
   "execution_count": 11
  },
  {
   "cell_type": "code",
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading training images\n",
      "Found 210000 files belonging to 2 classes.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-04-23 11:32:07.638708: I tensorflow/tsl/platform/default/subprocess.cc:304] Start cannot spawn child process: No such file or directory\n",
      "2024-04-23 11:32:07.641917: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.641943: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.641961: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.641983: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.642006: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.642048: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.642089: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.642452: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.643895: W tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:231] Falling back to the CUDA driver for PTX compilation; ptxas does not support CC 8.9\n",
      "2024-04-23 11:32:07.643911: W tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:234] Used ptxas at /usr/local/cuda/bin/ptxas\n",
      "2024-04-23 11:32:07.643978: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.644071: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.644121: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.644162: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.644199: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.644231: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.644277: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.644322: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.977218: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.977259: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.978078: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.978867: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.979398: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.979735: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.979944: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.980909: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.982034: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.982544: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.983165: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.984079: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.986483: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.987110: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.987536: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.988196: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:32:07.988658: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:32:07.989709: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading validation images\n",
      "Found 30000 files belonging to 2 classes.\n",
      "Loading test images\n",
      "Found 60000 files belonging to 2 classes.\n",
      "Class weights: {0: 1.5, 1: 0.75}\n"
     ]
    }
   ],
   "source": [
    "IMG_WIDTH=224\n",
    "IMG_HEIGHT=224\n",
    "BATCH_SIZE=128\n",
    "EPOCHS=100\n",
    "HIDDEN_UNITS=400\n",
    "TRAINABLE_BASE = False\n",
    "training_max_batches = None\n",
    "use_pooling = False\n",
    "use_validation_data_in_training = True\n",
    "use_data_aug = False\n",
    "\n",
    "real_train_path = \"/homeRepo/tanfoni/Dataset_merged/Train/Real/\"\n",
    "generated_train_path = \"/homeRepo/tanfoni/Dataset_merged/Train/Fake\"\n",
    "resize_dim = (IMG_WIDTH, IMG_HEIGHT)\n",
    "model_input_shape = (IMG_WIDTH, IMG_HEIGHT, 3)\n",
    "\n",
    "dataset_batch_size = BATCH_SIZE\n",
    "train_batch_size = int(dataset_batch_size * 0.6)\n",
    "validation_batch_size = int((dataset_batch_size - train_batch_size) * 0.6)\n",
    "test_batch_size = dataset_batch_size - train_batch_size - validation_batch_size\n",
    "\n",
    "parent_dir_train = \"/homeRepo/tanfoni/Dataset_merged/Train/\"\n",
    "real_dir_train = \"Real\"\n",
    "sint_dir_train = \"Fake\"\n",
    "\n",
    "parent_dir_val = \"/homeRepo/tanfoni/Dataset_merged/Valid/\"\n",
    "real_dir_val = \"Real\"\n",
    "sint_dir_val = \"Fake\"\n",
    "\n",
    "parent_dir_test = \"/homeRepo/tanfoni/Dataset_merged/Test/\"\n",
    "real_dir_test = \"Real\"\n",
    "sint_dir_test = \"Fake\"\n",
    "\n",
    "\n",
    "\n",
    "print(\"Loading training images\")\n",
    "train_dataset = tf.keras.preprocessing.image_dataset_from_directory(\n",
    "    parent_dir_train,\n",
    "    labels=\"inferred\",\n",
    "    class_names=[real_dir_train, sint_dir_train],\n",
    "    label_mode='binary',\n",
    "    color_mode='rgb',\n",
    "    batch_size=train_batch_size,\n",
    "    image_size=resize_dim,\n",
    "    shuffle=True,\n",
    "    seed=123,\n",
    ")\n",
    "\n",
    "if use_data_aug:\n",
    "    print(\"Performing data augmentation on training images\")\n",
    "    data_augmentation = ImageDataGenerator(\n",
    "        rotation_range=15,\n",
    "        width_shift_range=0.1,\n",
    "        height_shift_range=0.1,\n",
    "        label_mode='binary',\n",
    "        shear_range=0.2,\n",
    "        zoom_range=0.2,\n",
    "        horizontal_flip=True,\n",
    "        fill_mode='nearest'\n",
    "    )\n",
    "\n",
    "    train_dataset_aug = data_augmentation.flow_from_directory(\n",
    "        parent_dir_train,\n",
    "        target_size=resize_dim,\n",
    "        batch_size=train_batch_size,\n",
    "        label_mode='binary',\n",
    "        class_mode='binary',\n",
    "        shuffle=True,\n",
    "        seed=123\n",
    "    )\n",
    "\n",
    "\n",
    "print(\"Loading validation images\")\n",
    "val_dataset = tf.keras.preprocessing.image_dataset_from_directory(\n",
    "    parent_dir_val,\n",
    "    labels=\"inferred\",\n",
    "    class_names=[real_dir_val, sint_dir_val],\n",
    "    label_mode='binary',\n",
    "    color_mode='rgb',\n",
    "    batch_size=validation_batch_size,\n",
    "    image_size=resize_dim,\n",
    "    shuffle=True,\n",
    "    seed=123\n",
    ")\n",
    "\n",
    "print(\"Loading test images\")\n",
    "test_dataset = tf.keras.preprocessing.image_dataset_from_directory(\n",
    "    parent_dir_test,\n",
    "    labels=\"inferred\",\n",
    "    class_names=[real_dir_test, sint_dir_test],\n",
    "    label_mode='binary',\n",
    "    color_mode='rgb',\n",
    "    batch_size=test_batch_size,\n",
    "    image_size=resize_dim,\n",
    "    shuffle=True,\n",
    "    seed=123\n",
    ")\n",
    "\n",
    "# Balancing dataset\n",
    "n_fakes = len(os.listdir(os.path.join(parent_dir_train, sint_dir_train)))\n",
    "n_reals = len(os.listdir(os.path.join(parent_dir_train, real_dir_train)))\n",
    "total = n_fakes + n_reals\n",
    "weight_for_0 = (1 / n_reals) * total / 2.0\n",
    "weight_for_1 = (1 / n_fakes) * total / 2.0\n",
    "class_weights = {0: weight_for_0, 1: weight_for_1}\n",
    "print(\"Class weights: {}\".format(class_weights))\n"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-04-23T09:32:11.148287Z",
     "start_time": "2024-04-23T09:32:01.677073Z"
    }
   },
   "id": "f1b6a3feec368cbf",
   "execution_count": 12
  },
  {
   "cell_type": "code",
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_2\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                Output Shape                 Param #   Connected to                  \n",
      "==================================================================================================\n",
      " input_4 (InputLayer)        [(None, None, None, 3)]      0         []                            \n",
      "                                                                                                  \n",
      " tf.__operators__.getitem_4  (None, None, None, 3)        0         ['input_4[0][0]']             \n",
      "  (SlicingOpLambda)                                                                               \n",
      "                                                                                                  \n",
      " tf.nn.bias_add_4 (TFOpLamb  (None, None, None, 3)        0         ['tf.__operators__.getitem_4[0\n",
      " da)                                                                ][0]']                        \n",
      "                                                                                                  \n",
      " conv1_pad (ZeroPadding2D)   (None, None, None, 3)        0         ['tf.nn.bias_add_4[0][0]']    \n",
      "                                                                                                  \n",
      " conv1_conv (Conv2D)         (None, None, None, 64)       9472      ['conv1_pad[0][0]']           \n",
      "                                                                                                  \n",
      " conv1_bn (BatchNormalizati  (None, None, None, 64)       256       ['conv1_conv[0][0]']          \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv1_relu (Activation)     (None, None, None, 64)       0         ['conv1_bn[0][0]']            \n",
      "                                                                                                  \n",
      " pool1_pad (ZeroPadding2D)   (None, None, None, 64)       0         ['conv1_relu[0][0]']          \n",
      "                                                                                                  \n",
      " pool1_pool (MaxPooling2D)   (None, None, None, 64)       0         ['pool1_pad[0][0]']           \n",
      "                                                                                                  \n",
      " conv2_block1_1_conv (Conv2  (None, None, None, 64)       4160      ['pool1_pool[0][0]']          \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block1_1_bn (BatchNo  (None, None, None, 64)       256       ['conv2_block1_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block1_1_relu (Activ  (None, None, None, 64)       0         ['conv2_block1_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv2_block1_2_conv (Conv2  (None, None, None, 64)       36928     ['conv2_block1_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block1_2_bn (BatchNo  (None, None, None, 64)       256       ['conv2_block1_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block1_2_relu (Activ  (None, None, None, 64)       0         ['conv2_block1_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv2_block1_0_conv (Conv2  (None, None, None, 256)      16640     ['pool1_pool[0][0]']          \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block1_3_conv (Conv2  (None, None, None, 256)      16640     ['conv2_block1_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block1_0_bn (BatchNo  (None, None, None, 256)      1024      ['conv2_block1_0_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block1_3_bn (BatchNo  (None, None, None, 256)      1024      ['conv2_block1_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block1_add (Add)      (None, None, None, 256)      0         ['conv2_block1_0_bn[0][0]',   \n",
      "                                                                     'conv2_block1_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv2_block1_out (Activati  (None, None, None, 256)      0         ['conv2_block1_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv2_block2_1_conv (Conv2  (None, None, None, 64)       16448     ['conv2_block1_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block2_1_bn (BatchNo  (None, None, None, 64)       256       ['conv2_block2_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block2_1_relu (Activ  (None, None, None, 64)       0         ['conv2_block2_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv2_block2_2_conv (Conv2  (None, None, None, 64)       36928     ['conv2_block2_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block2_2_bn (BatchNo  (None, None, None, 64)       256       ['conv2_block2_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block2_2_relu (Activ  (None, None, None, 64)       0         ['conv2_block2_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv2_block2_3_conv (Conv2  (None, None, None, 256)      16640     ['conv2_block2_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block2_3_bn (BatchNo  (None, None, None, 256)      1024      ['conv2_block2_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block2_add (Add)      (None, None, None, 256)      0         ['conv2_block1_out[0][0]',    \n",
      "                                                                     'conv2_block2_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv2_block2_out (Activati  (None, None, None, 256)      0         ['conv2_block2_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv2_block3_1_conv (Conv2  (None, None, None, 64)       16448     ['conv2_block2_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block3_1_bn (BatchNo  (None, None, None, 64)       256       ['conv2_block3_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block3_1_relu (Activ  (None, None, None, 64)       0         ['conv2_block3_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv2_block3_2_conv (Conv2  (None, None, None, 64)       36928     ['conv2_block3_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block3_2_bn (BatchNo  (None, None, None, 64)       256       ['conv2_block3_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block3_2_relu (Activ  (None, None, None, 64)       0         ['conv2_block3_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv2_block3_3_conv (Conv2  (None, None, None, 256)      16640     ['conv2_block3_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv2_block3_3_bn (BatchNo  (None, None, None, 256)      1024      ['conv2_block3_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv2_block3_add (Add)      (None, None, None, 256)      0         ['conv2_block2_out[0][0]',    \n",
      "                                                                     'conv2_block3_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv2_block3_out (Activati  (None, None, None, 256)      0         ['conv2_block3_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv3_block1_1_conv (Conv2  (None, None, None, 128)      32896     ['conv2_block3_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block1_1_bn (BatchNo  (None, None, None, 128)      512       ['conv3_block1_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block1_1_relu (Activ  (None, None, None, 128)      0         ['conv3_block1_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv3_block1_2_conv (Conv2  (None, None, None, 128)      147584    ['conv3_block1_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block1_2_bn (BatchNo  (None, None, None, 128)      512       ['conv3_block1_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block1_2_relu (Activ  (None, None, None, 128)      0         ['conv3_block1_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv3_block1_0_conv (Conv2  (None, None, None, 512)      131584    ['conv2_block3_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block1_3_conv (Conv2  (None, None, None, 512)      66048     ['conv3_block1_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block1_0_bn (BatchNo  (None, None, None, 512)      2048      ['conv3_block1_0_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block1_3_bn (BatchNo  (None, None, None, 512)      2048      ['conv3_block1_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block1_add (Add)      (None, None, None, 512)      0         ['conv3_block1_0_bn[0][0]',   \n",
      "                                                                     'conv3_block1_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv3_block1_out (Activati  (None, None, None, 512)      0         ['conv3_block1_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv3_block2_1_conv (Conv2  (None, None, None, 128)      65664     ['conv3_block1_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block2_1_bn (BatchNo  (None, None, None, 128)      512       ['conv3_block2_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block2_1_relu (Activ  (None, None, None, 128)      0         ['conv3_block2_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv3_block2_2_conv (Conv2  (None, None, None, 128)      147584    ['conv3_block2_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block2_2_bn (BatchNo  (None, None, None, 128)      512       ['conv3_block2_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block2_2_relu (Activ  (None, None, None, 128)      0         ['conv3_block2_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv3_block2_3_conv (Conv2  (None, None, None, 512)      66048     ['conv3_block2_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block2_3_bn (BatchNo  (None, None, None, 512)      2048      ['conv3_block2_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block2_add (Add)      (None, None, None, 512)      0         ['conv3_block1_out[0][0]',    \n",
      "                                                                     'conv3_block2_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv3_block2_out (Activati  (None, None, None, 512)      0         ['conv3_block2_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv3_block3_1_conv (Conv2  (None, None, None, 128)      65664     ['conv3_block2_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block3_1_bn (BatchNo  (None, None, None, 128)      512       ['conv3_block3_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block3_1_relu (Activ  (None, None, None, 128)      0         ['conv3_block3_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv3_block3_2_conv (Conv2  (None, None, None, 128)      147584    ['conv3_block3_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block3_2_bn (BatchNo  (None, None, None, 128)      512       ['conv3_block3_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block3_2_relu (Activ  (None, None, None, 128)      0         ['conv3_block3_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv3_block3_3_conv (Conv2  (None, None, None, 512)      66048     ['conv3_block3_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block3_3_bn (BatchNo  (None, None, None, 512)      2048      ['conv3_block3_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block3_add (Add)      (None, None, None, 512)      0         ['conv3_block2_out[0][0]',    \n",
      "                                                                     'conv3_block3_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv3_block3_out (Activati  (None, None, None, 512)      0         ['conv3_block3_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv3_block4_1_conv (Conv2  (None, None, None, 128)      65664     ['conv3_block3_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block4_1_bn (BatchNo  (None, None, None, 128)      512       ['conv3_block4_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block4_1_relu (Activ  (None, None, None, 128)      0         ['conv3_block4_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv3_block4_2_conv (Conv2  (None, None, None, 128)      147584    ['conv3_block4_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block4_2_bn (BatchNo  (None, None, None, 128)      512       ['conv3_block4_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block4_2_relu (Activ  (None, None, None, 128)      0         ['conv3_block4_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv3_block4_3_conv (Conv2  (None, None, None, 512)      66048     ['conv3_block4_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv3_block4_3_bn (BatchNo  (None, None, None, 512)      2048      ['conv3_block4_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv3_block4_add (Add)      (None, None, None, 512)      0         ['conv3_block3_out[0][0]',    \n",
      "                                                                     'conv3_block4_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv3_block4_out (Activati  (None, None, None, 512)      0         ['conv3_block4_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv4_block1_1_conv (Conv2  (None, None, None, 256)      131328    ['conv3_block4_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block1_1_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block1_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block1_1_relu (Activ  (None, None, None, 256)      0         ['conv4_block1_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block1_2_conv (Conv2  (None, None, None, 256)      590080    ['conv4_block1_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block1_2_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block1_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block1_2_relu (Activ  (None, None, None, 256)      0         ['conv4_block1_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block1_0_conv (Conv2  (None, None, None, 1024)     525312    ['conv3_block4_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block1_3_conv (Conv2  (None, None, None, 1024)     263168    ['conv4_block1_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block1_0_bn (BatchNo  (None, None, None, 1024)     4096      ['conv4_block1_0_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block1_3_bn (BatchNo  (None, None, None, 1024)     4096      ['conv4_block1_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block1_add (Add)      (None, None, None, 1024)     0         ['conv4_block1_0_bn[0][0]',   \n",
      "                                                                     'conv4_block1_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv4_block1_out (Activati  (None, None, None, 1024)     0         ['conv4_block1_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv4_block2_1_conv (Conv2  (None, None, None, 256)      262400    ['conv4_block1_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block2_1_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block2_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block2_1_relu (Activ  (None, None, None, 256)      0         ['conv4_block2_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block2_2_conv (Conv2  (None, None, None, 256)      590080    ['conv4_block2_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block2_2_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block2_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block2_2_relu (Activ  (None, None, None, 256)      0         ['conv4_block2_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block2_3_conv (Conv2  (None, None, None, 1024)     263168    ['conv4_block2_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block2_3_bn (BatchNo  (None, None, None, 1024)     4096      ['conv4_block2_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block2_add (Add)      (None, None, None, 1024)     0         ['conv4_block1_out[0][0]',    \n",
      "                                                                     'conv4_block2_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv4_block2_out (Activati  (None, None, None, 1024)     0         ['conv4_block2_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv4_block3_1_conv (Conv2  (None, None, None, 256)      262400    ['conv4_block2_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block3_1_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block3_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block3_1_relu (Activ  (None, None, None, 256)      0         ['conv4_block3_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block3_2_conv (Conv2  (None, None, None, 256)      590080    ['conv4_block3_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block3_2_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block3_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block3_2_relu (Activ  (None, None, None, 256)      0         ['conv4_block3_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block3_3_conv (Conv2  (None, None, None, 1024)     263168    ['conv4_block3_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block3_3_bn (BatchNo  (None, None, None, 1024)     4096      ['conv4_block3_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block3_add (Add)      (None, None, None, 1024)     0         ['conv4_block2_out[0][0]',    \n",
      "                                                                     'conv4_block3_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv4_block3_out (Activati  (None, None, None, 1024)     0         ['conv4_block3_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv4_block4_1_conv (Conv2  (None, None, None, 256)      262400    ['conv4_block3_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block4_1_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block4_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block4_1_relu (Activ  (None, None, None, 256)      0         ['conv4_block4_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block4_2_conv (Conv2  (None, None, None, 256)      590080    ['conv4_block4_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block4_2_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block4_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block4_2_relu (Activ  (None, None, None, 256)      0         ['conv4_block4_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block4_3_conv (Conv2  (None, None, None, 1024)     263168    ['conv4_block4_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block4_3_bn (BatchNo  (None, None, None, 1024)     4096      ['conv4_block4_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block4_add (Add)      (None, None, None, 1024)     0         ['conv4_block3_out[0][0]',    \n",
      "                                                                     'conv4_block4_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv4_block4_out (Activati  (None, None, None, 1024)     0         ['conv4_block4_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv4_block5_1_conv (Conv2  (None, None, None, 256)      262400    ['conv4_block4_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block5_1_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block5_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block5_1_relu (Activ  (None, None, None, 256)      0         ['conv4_block5_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block5_2_conv (Conv2  (None, None, None, 256)      590080    ['conv4_block5_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block5_2_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block5_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block5_2_relu (Activ  (None, None, None, 256)      0         ['conv4_block5_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block5_3_conv (Conv2  (None, None, None, 1024)     263168    ['conv4_block5_2_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block5_3_bn (BatchNo  (None, None, None, 1024)     4096      ['conv4_block5_3_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block5_add (Add)      (None, None, None, 1024)     0         ['conv4_block4_out[0][0]',    \n",
      "                                                                     'conv4_block5_3_bn[0][0]']   \n",
      "                                                                                                  \n",
      " conv4_block5_out (Activati  (None, None, None, 1024)     0         ['conv4_block5_add[0][0]']    \n",
      " on)                                                                                              \n",
      "                                                                                                  \n",
      " conv4_block6_1_conv (Conv2  (None, None, None, 256)      262400    ['conv4_block5_out[0][0]']    \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block6_1_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block6_1_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block6_1_relu (Activ  (None, None, None, 256)      0         ['conv4_block6_1_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " conv4_block6_2_conv (Conv2  (None, None, None, 256)      590080    ['conv4_block6_1_relu[0][0]'] \n",
      " D)                                                                                               \n",
      "                                                                                                  \n",
      " conv4_block6_2_bn (BatchNo  (None, None, None, 256)      1024      ['conv4_block6_2_conv[0][0]'] \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " conv4_block6_2_relu (Activ  (None, None, None, 256)      0         ['conv4_block6_2_bn[0][0]']   \n",
      " ation)                                                                                           \n",
      "                                                                                                  \n",
      " global_average_pooling2d_7  (None, 256)                  0         ['conv4_block6_2_relu[0][0]'] \n",
      "  (GlobalAveragePooling2D)                                                                        \n",
      "                                                                                                  \n",
      " reshape_3 (Reshape)         (None, 1, 1, 256)            0         ['global_average_pooling2d_7[0\n",
      "                                                                    ][0]']                        \n",
      "                                                                                                  \n",
      " conv2d_27 (Conv2D)          (None, 1, 1, 24)             6168      ['reshape_3[0][0]']           \n",
      "                                                                                                  \n",
      " batch_normalization_27 (Ba  (None, 1, 1, 24)             96        ['conv2d_27[0][0]']           \n",
      " tchNormalization)                                                                                \n",
      "                                                                                                  \n",
      " conv2d_28 (Conv2D)          (None, None, None, 128)      295040    ['conv4_block6_2_relu[0][0]'] \n",
      "                                                                                                  \n",
      " conv2d_29 (Conv2D)          (None, None, None, 128)      524416    ['conv4_block6_2_relu[0][0]'] \n",
      "                                                                                                  \n",
      " conv2d_30 (Conv2D)          (None, None, None, 128)      819328    ['conv4_block6_2_relu[0][0]'] \n",
      "                                                                                                  \n",
      " conv2d_31 (Conv2D)          (None, None, None, 128)      2097280   ['conv4_block6_2_relu[0][0]'] \n",
      "                                                                                                  \n",
      " dynamic_upsample_9 (Dynami  (None, None, None, 24)       0         ['batch_normalization_27[0][0]\n",
      " cUpsample)                                                         ',                            \n",
      "                                                                     'conv4_block6_2_relu[0][0]'] \n",
      "                                                                                                  \n",
      " batch_normalization_28 (Ba  (None, None, None, 128)      512       ['conv2d_28[0][0]']           \n",
      " tchNormalization)                                                                                \n",
      "                                                                                                  \n",
      " batch_normalization_29 (Ba  (None, None, None, 128)      512       ['conv2d_29[0][0]']           \n",
      " tchNormalization)                                                                                \n",
      "                                                                                                  \n",
      " batch_normalization_30 (Ba  (None, None, None, 128)      512       ['conv2d_30[0][0]']           \n",
      " tchNormalization)                                                                                \n",
      "                                                                                                  \n",
      " batch_normalization_31 (Ba  (None, None, None, 128)      512       ['conv2d_31[0][0]']           \n",
      " tchNormalization)                                                                                \n",
      "                                                                                                  \n",
      " concatenate_8 (Concatenate  (None, None, None, 536)      0         ['dynamic_upsample_9[0][0]',  \n",
      " )                                                                   'batch_normalization_28[0][0]\n",
      "                                                                    ',                            \n",
      "                                                                     'batch_normalization_29[0][0]\n",
      "                                                                    ',                            \n",
      "                                                                     'batch_normalization_30[0][0]\n",
      "                                                                    ',                            \n",
      "                                                                     'batch_normalization_31[0][0]\n",
      "                                                                    ']                            \n",
      "                                                                                                  \n",
      " conv2d_32 (Conv2D)          (None, None, None, 128)      68736     ['concatenate_8[0][0]']       \n",
      "                                                                                                  \n",
      " conv2d_33 (Conv2D)          (None, None, None, 24)       1560      ['conv2_block3_2_relu[0][0]'] \n",
      "                                                                                                  \n",
      " batch_normalization_32 (Ba  (None, None, None, 128)      512       ['conv2d_32[0][0]']           \n",
      " tchNormalization)                                                                                \n",
      "                                                                                                  \n",
      " batch_normalization_33 (Ba  (None, None, None, 24)       96        ['conv2d_33[0][0]']           \n",
      " tchNormalization)                                                                                \n",
      "                                                                                                  \n",
      " dynamic_upsample_10 (Dynam  (None, None, None, 128)      0         ['batch_normalization_32[0][0]\n",
      " icUpsample)                                                        ',                            \n",
      "                                                                     'batch_normalization_33[0][0]\n",
      "                                                                    ']                            \n",
      "                                                                                                  \n",
      " concatenate_9 (Concatenate  (None, None, None, 152)      0         ['dynamic_upsample_10[0][0]', \n",
      " )                                                                   'batch_normalization_33[0][0]\n",
      "                                                                    ']                            \n",
      "                                                                                                  \n",
      " conv2d_34 (Conv2D)          (None, None, None, 128)      175232    ['concatenate_9[0][0]']       \n",
      "                                                                                                  \n",
      " batch_normalization_34 (Ba  (None, None, None, 128)      512       ['conv2d_34[0][0]']           \n",
      " tchNormalization)                                                                                \n",
      "                                                                                                  \n",
      " conv2d_35 (Conv2D)          (None, None, None, 128)      147584    ['batch_normalization_34[0][0]\n",
      "                                                                    ']                            \n",
      "                                                                                                  \n",
      " batch_normalization_35 (Ba  (None, None, None, 128)      512       ['conv2d_35[0][0]']           \n",
      " tchNormalization)                                                                                \n",
      "                                                                                                  \n",
      " dynamic_upsample_11 (Dynam  (None, None, None, 128)      0         ['batch_normalization_35[0][0]\n",
      " icUpsample)                                                        ',                            \n",
      "                                                                     'input_4[0][0]']             \n",
      "                                                                                                  \n",
      " global_average_pooling2d_8  (None, 128)                  0         ['dynamic_upsample_11[0][0]'] \n",
      "  (GlobalAveragePooling2D)                                                                        \n",
      "                                                                                                  \n",
      " dense_4 (Dense)             (None, 50)                   6450      ['global_average_pooling2d_8[0\n",
      "                                                                    ][0]']                        \n",
      "                                                                                                  \n",
      " dense_5 (Dense)             (None, 25)                   1275      ['dense_4[0][0]']             \n",
      "                                                                                                  \n",
      " dense_6 (Dense)             (None, 1)                    26        ['dense_5[0][0]']             \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 12468791 (47.56 MB)\n",
      "Trainable params: 4144983 (15.81 MB)\n",
      "Non-trainable params: 8323808 (31.75 MB)\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_nicco.summary()"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-04-23T09:52:39.955423Z",
     "start_time": "2024-04-23T09:52:39.660507Z"
    }
   },
   "id": "1d09d70ad3c8e909",
   "execution_count": 26
  },
  {
   "cell_type": "code",
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-04-23 11:52:44.150708: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:52:44.152487: W tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc:191] Failed to compile generated PTX with ptxas. Falling back to compilation by driver.\n",
      "2024-04-23 11:52:44.347621: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:52:44.510178: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:52:44.614751: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n",
      "2024-04-23 11:52:44.736828: E tensorflow/compiler/xla/stream_executor/gpu/asm_compiler.cc:114] *** WARNING *** You are using ptxas 11.0.194, which is older than 11.1. ptxas before 11.1 is known to miscompile XLA code, leading to incorrect results or invalid-address errors.\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2764/2764 [==============================] - 300s 106ms/step - loss: 0.4753 - accuracy: 0.7205 - precision: 0.8971 - recall: 0.6560 - val_loss: 0.4860 - val_accuracy: 0.7637 - val_precision: 0.7850 - val_recall: 0.8890\n",
      "Epoch 2/2\n",
      "2764/2764 [==============================] - 293s 106ms/step - loss: 0.4409 - accuracy: 0.7506 - precision: 0.8978 - recall: 0.7063 - val_loss: 0.4087 - val_accuracy: 0.7937 - val_precision: 0.8292 - val_recall: 0.8697\n"
     ]
    },
    {
     "data": {
      "text/plain": "<keras.src.callbacks.History at 0x7f3a343fa9b0>"
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_nicco.compile(optimizer=keras.optimizers.Adam(learning_rate=0.01),\n",
    "                    loss=keras.losses.BinaryCrossentropy(),\n",
    "                    metrics=[\"accuracy\",\n",
    "                    keras.metrics.Precision(name='precision'),\n",
    "                    keras.metrics.Recall(name='recall')])\n",
    "model_nicco.fit(train_dataset, epochs=2, class_weight=class_weights, validation_data=val_dataset)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2024-04-23T10:02:33.696098Z",
     "start_time": "2024-04-23T09:52:40.097348Z"
    }
   },
   "id": "e8833a519d5104bf",
   "execution_count": 27
  },
  {
   "cell_type": "code",
   "outputs": [],
   "source": [],
   "metadata": {
    "collapsed": false
   },
   "id": "77237a294b0c5e08",
   "execution_count": null
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}