mnist_basic_test.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "%run mnist_basic_data.ipynb\n",
    "%run mnist_basic_model.ipynb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.models import load_model\n",
    "from matplotlib import patches\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def transform_coord(coord, cell):\n",
    "    cy = cell // GRID_SIZE\n",
    "    cx = cell % GRID_SIZE\n",
    "    w = coord[12] * screen[0]\n",
    "    h = coord[13] * screen[1]\n",
    "    x = (coord[10] + cx)*(screen[0] / GRID_SIZE ) - (w/2)\n",
    "    y = (coord[11] + cy)*(screen[1] / GRID_SIZE ) - (h/2)\n",
    "    return (x, y), w, h"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_results(batch_, pred):\n",
    "    fig, axList = plt.subplots(2,4)\n",
    "    fig.set_size_inches(12,8)\n",
    "    \n",
    "    batch = [0, 1]\n",
    "    batch[0] = batch_[0]\n",
    "    batch[1] = batch_[1].reshape(-1, GRID_SIZE * GRID_SIZE, 15)\n",
    "    pred = pred.reshape(-1, GRID_SIZE * GRID_SIZE, 15)\n",
    "    \n",
    "    for i, ax in enumerate(axList.flatten()):\n",
    "        ax.set_axis_off()\n",
    "        ax.imshow(batch[0][i].reshape(86,86), cmap='gray')\n",
    "        \n",
    "        conf = pred[i][..., 14]\n",
    "        sorted_ind = conf.argsort()\n",
    "        top3_ind = sorted_ind[-3:]\n",
    "        \n",
    "#         title = ''\n",
    "        for j in top3_ind:\n",
    "#             title += str(pred[i][j][:10].argmax()) +':- '+ str(pred[i][j][:10].max() * pred[i][j][-1]) + ' ' if batch[1][i][j][:10].argmax() != pred[i][j][:10].argmax() else str('ok ')\n",
    "            \n",
    "            c1, c2, c3 = transform_coord(pred[i][j], j)\n",
    "            rec = patches.Rectangle(c1, c2, c3, edgecolor='r', fill=False, linewidth=1)\n",
    "            ax.add_patch(rec)\n",
    "            ax.text(c1[0], c1[1]+6, str(pred[i][j][:10].argmax(axis=-1)), color='c', fontsize=12)\n",
    "        \n",
    "#         ax.set_title(title)\n",
    "        true_conf = batch[1][i][..., 14]\n",
    "        sorted_ind = true_conf.argsort()\n",
    "        top3_ind = sorted_ind[-3:]\n",
    "        \n",
    "        for j in top3_ind:\n",
    "            c1, c2, c3 = transform_coord(batch[1][i][j], j)\n",
    "            rec = patches.Rectangle(c1, c2, c3, edgecolor='g', fill=False, linewidth=1)\n",
    "            ax.add_patch(rec)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "train, val, test = load_data()\n",
    "x_test, y_test = test\n",
    "\n",
    "test_gen = DataGenerator(x_test, y_test, batch_size=8)\n",
    "\n",
    "model = load_model('mnist_basic_yolo.h5', compile=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile('Adam', custom_loss, ['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch = test_gen.gen_batch()\n",
    "pred = model.predict(batch[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAsMAAAGhCAYAAACEWLmOAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIuJZCoZFRERExLP+D52//qq8IcO0AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fd38a779c50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_results(batch, pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Failures:\n",
    "\n",
    "### Data Generation: Images\n",
    "Couldn't find a way to remove black background in mnist images in PIL.\n",
    "So, instead just used an np array.\n",
    "\n",
    "### Data Generation: BBox\n",
    "True Bounding boxes were totally incorrect because of confusing x-y axes and [y][x] indices\n",
    "\n",
    "### Data Generation: BBox\n",
    "Prediction were totally incorrect because each cell in the grid was predicting coordinates w.r.t to top-left corner of the image.\n",
    "\n",
    "### Finally\n",
    "By making coordinates inside a cell relative to top-left corner of each cell, results became appreciable. But mostly all predicted boxes are larger.\n",
    "\n",
    "#### Still not using correct loss function, nor using multiple anchor boxes & non-max suppression\n",
    "#### Don't know how to calculate mAP or any other metric (COCO).\n",
    "#### Take a look at Focal Loss, Congenerous Cosine Loss, etc.\n",
    "#### Try VAE for classification by reconstructing input and using sampled vector for classification"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Tf2",
   "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": "3.5.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}