{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<img width=\"800px\" src=\"../fidle/img/00-Fidle-header-01.svg\"></img>\n",
"\n",
"# <!-- TITLE --> [VAE6] - Variational AutoEncoder (VAE) with CelebA (small)\n",
"<!-- DESC --> VAE with a more fun and realistic dataset - small resolution and batchable\n",
"<!-- AUTHOR : Jean-Luc Parouty (CNRS/SIMaP) -->\n",
"\n",
"## Objectives :\n",
" - Build and train a VAE model with a large dataset in small resolution(>70 GB)\n",
" - Understanding a more advanced programming model with data generator\n",
"\n",
"The [CelebFaces Attributes Dataset (CelebA)](http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html) contains about 200,000 images (202599,218,178,3). \n",
"\n",
"## What we're going to do :\n",
"\n",
" - Defining a VAE model\n",
" - Build the model\n",
" - Train it\n",
" - Follow the learning process with Tensorboard\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 1 - Setup environment\n",
"### 1.1 - Python stuff"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import tensorflow as tf\n",
"import numpy as np\n",
"import os,sys\n",
"from importlib import reload\n",
"\n",
"import modules.vae\n",
"import modules.data_generator\n",
"\n",
"reload(modules.data_generator)\n",
"reload(modules.vae)\n",
"\n",
"from modules.vae import VariationalAutoencoder\n",
"from modules.data_generator import DataGenerator\n",
"\n",
"sys.path.append('..')\n",
"import fidle.pwk as ooo\n",
"reload(ooo)\n",
"\n",
"ooo.init()\n",
"\n",
"VariationalAutoencoder.about()\n",
"DataGenerator.about()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 1.2 - The good place"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"place, dataset_dir = ooo.good_place( { 'GRICAD' : f'{os.getenv(\"SCRATCH_DIR\",\"\")}/PROJECTS/pr-fidle/datasets/celeba',\n",
" 'IDRIS' : f'{os.getenv(\"WORK\",\"\")}/datasets/celeba',\n",
" 'HOME' : f'{os.getenv(\"HOME\",\"\")}/datasets/celeba'} )\n",
"\n",
"# ---- train/test datasets\n",
"\n",
"train_dir = f'{dataset_dir}/clusters.train'\n",
"test_dir = f'{dataset_dir}/clusters.test'"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 2 - DataGenerator and validation data\n",
"Ok, everything's perfect, now let's instantiate our generator for the entire dataset."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"data_gen = DataGenerator(train_dir, 32, k_size=1)\n",
"x_test = np.load(f'{test_dir}/images-000.npy')\n",
"\n",
"print(f'Data generator : {len(data_gen)} batchs of {data_gen.batch_size} images, or {data_gen.dataset_size} images')\n",
"print(f'x_test : {len(x_test)} images')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 3 - Get VAE model"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"tag = f'CelebA.006-S.{os.getenv(\"SLURM_JOB_ID\",\"unknown\")}'\n",
"\n",
"input_shape = (128, 128, 3)\n",
"z_dim = 200\n",
"verbose = 1\n",
"\n",
"encoder= [ {'type':'Conv2D', 'filters':32, 'kernel_size':(3,3), 'strides':2, 'padding':'same', 'activation':'relu'},\n",
" {'type':'Dropout', 'rate':0.25},\n",
" {'type':'Conv2D', 'filters':64, 'kernel_size':(3,3), 'strides':2, 'padding':'same', 'activation':'relu'},\n",
" {'type':'Dropout', 'rate':0.25},\n",
" {'type':'Conv2D', 'filters':64, 'kernel_size':(3,3), 'strides':2, 'padding':'same', 'activation':'relu'},\n",
" {'type':'Dropout', 'rate':0.25},\n",
" {'type':'Conv2D', 'filters':64, 'kernel_size':(3,3), 'strides':2, 'padding':'same', 'activation':'relu'},\n",
" {'type':'Dropout', 'rate':0.25},\n",
" ]\n",
"\n",
"decoder= [ {'type':'Conv2DTranspose', 'filters':64, 'kernel_size':(3,3), 'strides':2, 'padding':'same', 'activation':'relu'},\n",
" {'type':'Dropout', 'rate':0.25},\n",
" {'type':'Conv2DTranspose', 'filters':64, 'kernel_size':(3,3), 'strides':2, 'padding':'same', 'activation':'relu'},\n",
" {'type':'Dropout', 'rate':0.25},\n",
" {'type':'Conv2DTranspose', 'filters':32, 'kernel_size':(3,3), 'strides':2, 'padding':'same', 'activation':'relu'},\n",
" {'type':'Dropout', 'rate':0.25},\n",
" {'type':'Conv2DTranspose', 'filters':3, 'kernel_size':(3,3), 'strides':2, 'padding':'same', 'activation':'sigmoid'}\n",
" ]\n",
"\n",
"vae = modules.vae.VariationalAutoencoder(input_shape = input_shape, \n",
" encoder_layers = encoder, \n",
" decoder_layers = decoder,\n",
" z_dim = z_dim, \n",
" verbose = verbose,\n",
" run_tag = tag)\n",
"vae.save(model=None)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 4 - Compile it"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"optimizer = tf.keras.optimizers.Adam(1e-4)\n",
"# optimizer = 'adam'\n",
"r_loss_factor = 10000\n",
"\n",
"vae.compile(optimizer, r_loss_factor)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 5 - Train\n",
"For 10 epochs, adam optimizer : \n",
"- Run time at IDRIS : 1299.77 sec. - 0:21:39\n",
"- Run time at GRICAD : 2092.77 sec. - 0:34:52"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"epochs = 10\n",
"initial_epoch = 0"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"vae.train(data_generator = data_gen,\n",
" x_test = x_test,\n",
" epochs = epochs,\n",
" initial_epoch = initial_epoch\n",
" )"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"---\n",
"<img width=\"80px\" src=\"../fidle/img/00-Fidle-logo-01.svg\"></img>"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.6"
}
},
"nbformat": 4,
"nbformat_minor": 4
}