op_test.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import tensorflow.contrib.slim as slim\n",
    "# NOTE: make sure to build the ops\n",
    "import filter_ops\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "import pickle\n",
    "import numpy as np\n",
    "import timeit\n",
    "\n",
    "import os\n",
    "os.environ['CUDA_VISIBLE_DEVICES'] = '1'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def compute_spatial(height, width):\n",
    "    # TODO: is this a bottleneck?\n",
    "    xs = tf.tile(tf.reshape(tf.range(0, width), [1, width, 1]), [height, 1, 1])\n",
    "    ys = tf.tile(tf.reshape(tf.range(0, height), [height, 1, 1]), [1, width, 1])\n",
    "    return tf.cast(tf.concat([xs, ys], 2), tf.float32)\n",
    "\n",
    "def meanfield_op(unary, \n",
    "                 features_sp, features_bl, \n",
    "                 kernel_sp, kernel_bl, kernel_compat,\n",
    "                 num_iters=10, step=1.0, norm_eps=1e-20, \n",
    "                 name='meanfield'):\n",
    "\n",
    "    with tf.variable_scope(name):\n",
    "        N = unary.shape[0]\n",
    "\n",
    "        norm_feed = tf.ones([N,1], dtype=tf.float32)\n",
    "\n",
    "        # TODO: should we make a batch version?\n",
    "        lattice_sp = filter_ops.permuto_init(features_sp)\n",
    "        norm_sp = 1.0 / (norm_eps + filter_ops.ph_filter(norm_feed, lattice_sp))\n",
    "\n",
    "        lattice_bl = filter_ops.permuto_init(features_bl)\n",
    "        norm_bl = 1.0 / (norm_eps + filter_ops.ph_filter(norm_feed, lattice_bl))\n",
    "\n",
    "        def _meanfield_iter(prev, it):\n",
    "            prob = slim.softmax(prev)\n",
    "            \n",
    "            msg_sp_raw = filter_ops.ph_filter(prob, lattice_sp)\n",
    "            msg_sp = tf.matmul(msg_sp_raw * norm_sp, kernel_sp)\n",
    "            \n",
    "            msg_bl_raw = filter_ops.ph_filter(prob, lattice_bl)\n",
    "            msg_bl = tf.matmul(msg_bl_raw * norm_bl, kernel_bl)\n",
    "            # TODO: we can even weight these between each other\n",
    "            pairwise = tf.matmul(msg_sp + msg_bl, kernel_compat)\n",
    "            nat_up = unary - pairwise\n",
    "            return (1.0 - step) * prev + step * nat_up\n",
    "\n",
    "        nats = tf.foldl(_meanfield_iter, \n",
    "                        tf.range(num_iters), \n",
    "                        unary)\n",
    "        return nats"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# loading inputs\n",
    "in_unaries = np.load('data/input.unaries.npy')\n",
    "in_unaries = np.squeeze(in_unaries).transpose((1,2,0))\n",
    "in_rgb = np.load('data/input.bgr.npy')[:,:,::-1]\n",
    "\n",
    "OH, OW, NC = in_unaries.shape  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "building the graph\n",
      "done!\n"
     ]
    }
   ],
   "source": [
    "# building the inference\n",
    "print(f'building the graph')\n",
    "with tf.variable_scope('inference'):\n",
    "    height, width, num_classes = OH, OW, NC\n",
    "\n",
    "    # TODO: the last dimension can change actually\n",
    "    unary = tf.placeholder(tf.float32, [height*width, num_classes], name='unary')\n",
    "    rgb = tf.placeholder(tf.float32, [height, width, 3], name='rgb')\n",
    "\n",
    "    # various parameters\n",
    "    sxy_sp = tf.placeholder(tf.float32, [])\n",
    "    sxy_bl = tf.placeholder(tf.float32, [])\n",
    "    srgb_bl = tf.placeholder(tf.float32, [])\n",
    "\n",
    "    # TODO: the kernels\n",
    "    kernel_sp = 1.0 * tf.diag(tf.ones(num_classes, dtype=np.float32))\n",
    "    kernel_bl = 3.0 * tf.diag(tf.ones(num_classes, dtype=np.float32))\n",
    "    kernel_compat = -1.0 * tf.diag(tf.ones(num_classes, dtype=np.float32))\n",
    "\n",
    "    # computing features\n",
    "    features_sp = tf.reshape(compute_spatial(height, width) / sxy_sp, \n",
    "                             [height*width, 2])\n",
    "    features_bl = tf.reshape(tf.concat(axis=2, \n",
    "                                       values=[compute_spatial(height, width) / sxy_bl,\n",
    "                                         rgb / srgb_bl]),\n",
    "                           [height*width, 2+3])\n",
    "\n",
    "    nats = meanfield_op(unary, \n",
    "                        features_sp, features_bl, \n",
    "                        kernel_sp, kernel_bl, kernel_compat, \n",
    "                        num_iters=20)\n",
    "    probs = slim.softmax(nats)\n",
    "    probs = tf.reshape(probs, [OH,OW,NC])\n",
    "print(f'done!')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "elapsed: 3.6725768940523267\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7f79154224e0>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUEAAAEyCAYAAABkuGJeAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n/wFG5rVDI1jn1wAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f791c0670b8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "with tf.Session() as sess:\n",
    "    feed_dict = {\n",
    "        unary : in_unaries.reshape((OH*OW, NC)),\n",
    "        rgb : in_rgb,\n",
    "        sxy_sp : 1.0,\n",
    "        sxy_bl : 3.0,\n",
    "        srgb_bl : 60.0\n",
    "    }\n",
    "\n",
    "    start_time = timeit.default_timer()\n",
    "    _probs = sess.run(probs, feed_dict)\n",
    "    print(f'elapsed: {timeit.default_timer() - start_time}')\n",
    "    _mask = np.argmax(_probs, axis=2)\n",
    "    \n",
    "plt.figure(figsize=(5, 5))\n",
    "plt.imshow(_mask)  \n",
    "    "
   ]
  }
 ],
 "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}