EXAMPLE OF VIDEO-QA IMPLEMENTATION WITH TENSORFLOW 1.2RC0

VIDEO-QA with Tensorflow 1.2RC0.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "EFFECTIVE TENSORFLOW FOR NON-EXPERTS https://www.youtube.com/watch?v=5DknTFbcGVM"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**You could download the latest version of Tensorflow, all branches here :**\n",
    "\n",
    "==> **https://www.tensorflow.org/versions/**\n",
    "\n",
    "==> ** https://pypi.python.org/pypi/tensorflow#downloads** (Installation : ***pip install package_name***)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "### EXAMPLE OF VIDEO-QA IMPLEMENTATION WITH TENSORFLOW 1.2RC0\n",
    "### SAME SOURCE ON KERAS BLOG : https://keras.io/getting-started/functional-api-guide/#shared-layers\n",
    "### Video question answering model part\n",
    "from __future__ import print_function\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.gridspec as gridspec\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "print(\"tf.__version__ : \", tf.__version__)\n",
    "\n",
    "from tensorflow import keras\n",
    "from tensorflow.keras import layers\n",
    "from tensorflow.keras.applications import InceptionV3\n",
    "\n",
    "### Turning frames into a vector, with pre-trained representations\n",
    "video = keras.Input(shape=(None, 150, 150, 3), name='video')\n",
    "cnn = InceptionV3(weights='imagenet',\n",
    "                  include_top=False,\n",
    "                  pooling='avg')\n",
    "cnn.trainable = False\n",
    "frame_features = layers.TimeDistributed(cnn)(video)\n",
    "video_vector = layers.LSTM(256)(frame_features)\n",
    "\n",
    "### Turning a sequence of words into a vector\n",
    "question = keras.Input(shape=(None, ), dtype='int32', name='question')\n",
    "embedded_words = layers.Embedding(input_voc_size, 256)(question)\n",
    "question_vector = layers.LSTM(128)(embedded_words)\n",
    "\n",
    "### Predicting an answer word\n",
    "x = layers.concatenate([video_vector, question_vector])\n",
    "x = layers.dense(128, activation=tf.nn.relu)(x)\n",
    "predictions = layers.dense(output_voc_size, name='predictions')(x)\n",
    "\n",
    "### Setting up the training configuration\n",
    "model = keras.models.Model([video, question], predictions)\n",
    "model.compile(optimizer=tf.AdamOptimizer(),\n",
    "              loss=tf.softmax_crossentropy_with_logits)\n",
    "\n",
    "\n",
    "### Leveraging Experiment for distributed training\n",
    "def experiment_fn(config, params):\n",
    "    model = ...\n",
    "    estimator = model.get_estimator(config=...)\n",
    "    return Experiment(estimator,\n",
    "                      train_input_fn=pandas_input_fn(...),\n",
    "                      eval_input_fn=pandas_input_fn(...))\n",
    "\n",
    "\n",
    "if __name__ == '__main__':\n",
    "    tf.contrib.train.run_experiment(experiment_fn)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.5.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}