{
"metadata": {
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.4"
},
"orig_nbformat": 2,
"kernelspec": {
"name": "python394jvsc74a57bd0d55a872fb12b64c3eb6a530d12935ddebcb38da0925d2cc3bd9c2ebc1d370b0d",
"display_name": "Python 3.9.4 64-bit"
},
"metadata": {
"interpreter": {
"hash": "d55a872fb12b64c3eb6a530d12935ddebcb38da0925d2cc3bd9c2ebc1d370b0d"
}
}
},
"nbformat": 4,
"nbformat_minor": 2,
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"INFO:tensorflow:Enabling eager execution\n",
"INFO:tensorflow:Enabling v2 tensorshape\n",
"INFO:tensorflow:Enabling resource variables\n",
"INFO:tensorflow:Enabling tensor equality\n",
"INFO:tensorflow:Enabling control flow v2\n",
"WARNING:tensorflow:SavedModel saved prior to TF 2.5 detected when loading Keras model. Please ensure that you are saving the model with model.save() or tf.keras.models.save_model(), *NOT* tf.saved_model.save(). To confirm, there should be a file named \"keras_metadata.pb\" in the SavedModel directory.\n",
"Model used: firstModel\n"
]
}
],
"source": [
"#@title Imports\n",
"#%load_ext autoreload #Need to uncomment for import sometime, dont understand\n",
"\n",
"#Tensorflow :\n",
"import tensorflow as tf\n",
"from tensorflow import keras\n",
"from tensorflow.keras import datasets, layers, models, losses\n",
"import tensorflow_datasets as tfds\n",
"#from google.colab import files\n",
"\n",
"#Others :\n",
"from matplotlib import image\n",
"import os\n",
"import numpy as np\n",
"import matplotlib.pyplot as plt\n",
"import matplotlib\n",
"import random as rd\n",
"import cv2\n",
"import csv\n",
"\n",
"#Data loaders :\n",
"from loadFer2013DS import *\n",
"from loadRavdessDS import *\n",
"from loadExpWDS import *\n",
"from loadAffwildDS import *\n",
"\n",
"#Others\n",
"from utils import *\n",
"from config import *"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#Chargement des données\n",
"print(\"Array loading...\")\n",
"Xf = np.load(\"data/array/Xf\")\n",
"Xe = np.load(\"data/array/Xe\")\n",
"Xa = np.load(\"data/array/Xa\")\n",
"Xr = np.load(\"data/array/Xr\")\n",
"\n",
"Yf = np.load(\"data/array/Yf\")\n",
"Ye = np.load(\"data/array/Ye\")\n",
"Ya = np.load(\"data/array/Ya\")\n",
"Yr = np.load(\"data/array/Yr\")\n",
"\n",
"print(\"Concatenation...\")\n",
"X = np.concatenate([Xf, Xa, Xe, Xr])\n",
"Y = np.concatenate([Yf, Ya, Ye, Yr])"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#Enregistre X et Y directement, à faire si assez de ram\n",
"np.save(\"data/array/X\", X)\n",
"np.save(\"data/array/Y\", Y)\n",
"def loadData():\n",
" return np.load(\"data/array/X\"), np.load(\"data/array/Y\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#@title Visualisation de chaque dataset\n",
"for X, Y, name in zip([Xf, Xr, Xe, Xa], [Yf, Yr, Ye, Ya], [\"fer2013\", \"ravdess\", \"expW\", \"affwild\"]):\n",
" N=5\n",
" M=5\n",
" print(\"Dataset:\", name)\n",
" print(\"Images:\", X.shape, \"La bels:\", Y.shape)\n",
" plt.figure()\n",
" for i in range(N*M):\n",
" if X.shape[0] == 0: continue\n",
" k = rd.randrange(X.shape[0])\n",
" plt.subplot(N, M, i+1)\n",
" plt.xticks([])\n",
" plt.yticks([])\n",
" plt.grid(False)\n",
"\n",
" afficher(X[k])\n",
" plt.title(emotions[int(Y[k])])\n",
" plt.show()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#Visualisation du dataset global\n",
"print(\"X_train:\", X_train.shape)\n",
"print(\"Y_train:\", Y_train.shape)\n",
"\n",
"print(\"\\nX_test:\", X_test.shape)\n",
"print(\"Y_test:\", Y_test.shape)\n",
"\n",
"N=5\n",
"M=5\n",
"plt.figure()\n",
"for i in range(N*M):\n",
" k = rd.randrange(X_train.shape[0])\n",
" plt.subplot(N, M, i+1)\n",
" plt.xticks([])\n",
" plt.yticks([])\n",
" plt.grid(False)\n",
"\n",
" afficher(X_train[k])\n",
" plt.title(emotions[int(Y_train[k])])\n",
"plt.show()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#@title Hyperparamètres\n",
"epochs = 2\n",
"batch_size = 128\n",
"validation_size = 0.1"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#Labels catégoriques\n",
"Ycat = keras.utils.to_categorical(Y)\n",
"\n",
"print(\"X\", X.shape)\n",
"print(\"Y\", Ycat.shape)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#MODELE\n",
"class MyModel(keras.Sequential):\n",
"\n",
" def __init__(self, input_shape):\n",
" super(MyModel, self).__init__()\n",
" #Pre processing\n",
" self.add(keras.layers.experimental.preprocessing.RandomContrast(factor=(0.5,0.5)))\n",
" self.add(keras.layers.experimental.preprocessing.RandomFlip(mode=\"horizontal\"))\n",
" \n",
" #48*48 *1\n",
" self.add(keras.layers.Conv2D(32, kernel_size = (3, 3), activation = 'relu', input_shape = input_shape)) \n",
" self.add(keras.layers.MaxPooling2D(pool_size = 2))\n",
" self.add(keras.layers.BatchNormalization())\n",
"\n",
" #23*23 *32\n",
" self.add(keras.layers.Conv2D(64, kernel_size = (3, 3), activation = 'relu'))\n",
" self.add(keras.layers.MaxPooling2D(pool_size = 2))\n",
" self.add(keras.layers.BatchNormalization())\n",
"\n",
" #10*10 *64\n",
" self.add(keras.layers.Conv2D(128, kernel_size = (3, 3), activation = 'relu'))\n",
" self.add(keras.layers.MaxPooling2D(pool_size = 2))\n",
" self.add(keras.layers.BatchNormalization())\n",
"\n",
" #4*4 *128\n",
" self.add(keras.layers.Conv2D(256, kernel_size = (3, 3), activation = 'relu'))\n",
" self.add(keras.layers.MaxPooling2D(pool_size = 2))\n",
" self.add(keras.layers.BatchNormalization())\n",
"\n",
" #1*1 *256\n",
" self.add(keras.layers.Flatten())\n",
" self.add(keras.layers.Dense(128, activation = 'relu'))\n",
" self.add(keras.layers.Dropout(0.2))\n",
" self.add(keras.layers.Dense(64, activation = 'relu'))\n",
" self.add(keras.layers.Dropout(0.2))\n",
" #self.add(keras.layers.BatchNormalization())\n",
" self.add(keras.layers.Dense(7, activation = 'softmax'))\n",
" #7\n",
" \n",
" def predir(self, monImage):\n",
" return self.predict(np.array([monImage]))[0,:]\n",
"\n",
" def compile_o(self):\n",
" self.compile(optimizer = 'adam', loss=losses.categorical_crossentropy, metrics = ['accuracy'])\n",
"\n",
"myModel = MyModel(input_shape)\n",
"myModel.compile_o()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"theImage = X_train[0]\n",
"afficher(theImage)\n",
"print(predir(myModel, theImage))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"history = myModel.fit(X, Y, epochs=5, validation_rate=0.05)\n",
"\n",
"#Affichage de l'historique de l'apprentissage\n",
"plt.plot(history.history['accuracy'], label='accuracy')\n",
"plt.plot(history.history['val_accuracy'], label='val_accuracy')\n",
"plt.legend()\n",
"plt.ylim([min(history.history['val_accuracy']+history.history['accuracy']), 1])\n",
"plt.show()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"myModel.save('exp904')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
]
}