2018-05-03-SmallNORB.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SmallNORB\n",
    "\n",
    "SmallNORB is a dataset used in machine learning. It shows toys from various angles and in different lighting.\n",
    "\n",
    "It's available here: https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/\n",
    "\n",
    "This post:\n",
    " - downloads the files\n",
    " - parses the files' headers and loads the data into numpy arrays\n",
    " - draws a few of the images and has an interactive tool.\n",
    "\n",
    "I've only tested this on Python 3.5 on a Mac. Other than that, it should only require numpy, jupyter, and matplotlib.\n",
    "\n",
    "You might also be interested in https://github.com/ndrplz/small_norb/ which also helped me figure out how to do this in Python.\n",
    "\n",
    "This post is available as a [blog post](http://jessicastringham.net/2018/05/03/smallnorb.html) or a [notebook](https://github.com/jessstringham/notebooks/blob/master/2018-05-03-SmallNORB.ipynb)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Overview of files\n",
    "\n",
    "Before I start, here's an overview of the dataset based on information from [the page (the source of truth for this dataset!)](https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/)\n",
    "\n",
    "There are a number of toys of different categories. Each toy is photographed from a fixed number of angles and lighting. The toys are grouped into one of 5 categories such as 'cars'. In each category, there are 10 toys which are referred to as an \"instances.\" \n",
    "\n",
    "There are two sets of files, one for training and one for testing. The training and testing datasets are divided across instances. So the 24,300 examples of instances 4, 6, 7, 8, 9 of all classes make up the training set. \n",
    "\n",
    "Each set contains three files. Each file's first dimension represents the 24,300 examples. The files are:\n",
    "\n",
    " - **`dat`**: The images! An example is of size `(2, 96, 96)`. This represents two `(96, 96)` images, one for each of two cameras.\n",
    " - **`cat`**: An example is an int. This represents the category id.\n",
    " - **`info`**: An example is `(4,)`. This represents the metadata about lighting/angle/etc. Specifically:\n",
    "   - **`instance`**: id `[0-9]` of which toy from the category it is.\n",
    "   - **`elevation`**: id `[0-8]` representing the elevation of the camera (from 30 to 70 using increments of 5).\n",
    "   - **`azimuth`**: id `[0-34]` (only even numbers!), where multiplying by 10 represents the azimuth. Heads up that instances aren't aligned! `azimuth=0` for two toys will point in two different directions.\n",
    "   - **`lighting`**: id `[0-5]` representing the lighting."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from collections import namedtuple\n",
    "import gzip\n",
    "import os\n",
    "import requests\n",
    "import struct\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Downloading\n",
    "\n",
    "The next code block downloads and decompresses the 6 files from https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/ in the `DATA_FOLDER` location.\n",
    "If you already have the decompressed files, you can point `DATA_FOLDER` to their folder.\n",
    "\n",
    "*Heads up, the `dat` files are large. This may take a while.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "skipping data/smallnorb-5x46789x9x18x6x2x96x96-training-info.mat, already downloaded\n",
      "skipping data/smallnorb-5x46789x9x18x6x2x96x96-training-cat.mat, already downloaded\n",
      "skipping data/smallnorb-5x46789x9x18x6x2x96x96-training-dat.mat, already downloaded\n",
      "skipping data/smallnorb-5x01235x9x18x6x2x96x96-testing-info.mat, already downloaded\n",
      "skipping data/smallnorb-5x01235x9x18x6x2x96x96-testing-cat.mat, already downloaded\n",
      "skipping data/smallnorb-5x01235x9x18x6x2x96x96-testing-dat.mat, already downloaded\n"
     ]
    }
   ],
   "source": [
    "DATA_FOLDER = 'data'\n",
    "URL = 'https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/'\n",
    "\n",
    "PREFIXES = {\n",
    "    'train': 'smallnorb-5x46789x9x18x6x2x96x96-training-',\n",
    "    'test': 'smallnorb-5x01235x9x18x6x2x96x96-testing-',\n",
    "}\n",
    "\n",
    "FILE_TYPES = ['info', 'cat', 'dat']\n",
    "\n",
    "SUFFIX = '.mat'\n",
    "\n",
    "for prefix in PREFIXES.values():\n",
    "    for file_type in FILE_TYPES:\n",
    "        file_name = prefix + file_type + SUFFIX\n",
    "        output_location = os.path.join(DATA_FOLDER, file_name)\n",
    "        file_location = URL + file_name + '.gz'\n",
    "        \n",
    "        if os.path.exists(output_location):\n",
    "            print('skipping {}, already downloaded'.format(output_location))\n",
    "            continue\n",
    "        \n",
    "        print('Downloading {} to {}'.format(file_location, output_location))\n",
    "\n",
    "        r = requests.get(file_location)\n",
    "        with open(output_location, 'wb') as f:\n",
    "            f.write(gzip.decompress(r.content))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Reading the files\n",
    "\n",
    "[The site](https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/) walks through how the files are represented.\n",
    "Each of the files has a header that explains how to read the data structure of the rest of the file. \n",
    "\n",
    "In this block, for each file, I read the header, then use it to load a numpy array of the file."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Reading smallnorb-5x46789x9x18x6x2x96x96-training-info.mat\n",
      "dtype <class 'numpy.int32'>\n",
      "shape [24300, 4]\n",
      "Reading smallnorb-5x46789x9x18x6x2x96x96-training-cat.mat\n",
      "dtype <class 'numpy.int32'>\n",
      "shape [24300]\n",
      "Reading smallnorb-5x46789x9x18x6x2x96x96-training-dat.mat\n",
      "dtype <class 'numpy.uint8'>\n",
      "shape [24300, 2, 96, 96]\n",
      "Reading smallnorb-5x01235x9x18x6x2x96x96-testing-info.mat\n",
      "dtype <class 'numpy.int32'>\n",
      "shape [24300, 4]\n",
      "Reading smallnorb-5x01235x9x18x6x2x96x96-testing-cat.mat\n",
      "dtype <class 'numpy.int32'>\n",
      "shape [24300]\n",
      "Reading smallnorb-5x01235x9x18x6x2x96x96-testing-dat.mat\n",
      "dtype <class 'numpy.uint8'>\n",
      "shape [24300, 2, 96, 96]\n"
     ]
    }
   ],
   "source": [
    "# helper function to read int from file\n",
    "def read_int(f):\n",
    "    num, = struct.unpack('i', f.read(4))\n",
    "    return num\n",
    "\n",
    "\n",
    "# From https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/ \n",
    "# \"The magic number encodes the element type of the matrix\"\n",
    "# Note: I only copied over the ones I needed for these files.\n",
    "map_magic_number_to_data_type = {\n",
    "    '1e3d4c55': np.uint8,\n",
    "    '1e3d4c54': np.int32,\n",
    "}\n",
    "\n",
    "loaded_data = {}\n",
    "\n",
    "for dataset, prefix in PREFIXES.items():\n",
    "    for filetype in FILE_TYPES:\n",
    "        filename = prefix + filetype + SUFFIX\n",
    "        print('Reading {}'.format(filename))\n",
    "        \n",
    "        file_loc = os.path.join(DATA_FOLDER, filename)\n",
    "        with open(file_loc, 'rb') as f:\n",
    "            # Read the magic_num, convert it to hexadecimal, and look up the data_type\n",
    "            raw_magic_num = read_int(f)\n",
    "            magic_num = format(raw_magic_num, '02x')\n",
    "            data_type = map_magic_number_to_data_type[magic_num]\n",
    "            print('dtype', data_type)\n",
    "\n",
    "            # Read how many dimensions to expect\n",
    "            ndim = read_int(f)\n",
    "            \n",
    "            # Read at least 3 ints, or however many ndim there are\n",
    "            shape = [\n",
    "                read_int(f)\n",
    "                for i in range(max(ndim, 3))\n",
    "            ]   \n",
    "            # But in case ndims < 3, take at most n_dim elements\n",
    "            shape = shape[:ndim]\n",
    "            print('shape', shape)\n",
    "    \n",
    "            # Now load the actual data!\n",
    "            loaded_data[(dataset, filetype)] = np.fromfile(\n",
    "                f, \n",
    "                dtype=data_type, \n",
    "                count=np.prod(shape)\n",
    "            ).reshape(shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Viewing an image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nXf8J41ntAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x10d8781d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(loaded_data[('train', 'dat')][0, 0], cmap='gray')\n",
    "plt.axis('off')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Saving the data in a new format\n",
    "\n",
    "If at this point, you want SmallNORB in `npz` file for another project, you could do something like:\n",
    "\n",
    "    np.savez_compressed(\n",
    "        'data/smallnorb-20180503-train.npz', \n",
    "        inputs=loaded_data[('train', 'dat'), \n",
    "        targets=loaded_data[('train', 'cat')],\n",
    "    )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Exploring the dataset\n",
    "\n",
    "Now I'll explore the dataset.\n",
    "\n",
    "The downloaded dataset splits instances across the train and test set. For exploration, I'm going to concatenate them and make a mapping from metadata and category to the image index."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# I'll index examples out of this dataset\n",
    "all_dat = np.vstack((\n",
    "    loaded_data[('train', 'dat')], \n",
    "    loaded_data[('test', 'dat')]\n",
    "))\n",
    "\n",
    "\n",
    "# These I'll use to build the example_lookup table\n",
    "all_info = np.vstack((\n",
    "    loaded_data[('train', 'info')], \n",
    "    loaded_data[('test', 'info')]\n",
    "))\n",
    "# hstack these because it's a 1d vec\n",
    "all_targets = np.hstack((\n",
    "    loaded_data[('train', 'cat')], \n",
    "    loaded_data[('test', 'cat')]\n",
    "))\n",
    "\n",
    "# I'll concatenate the category and the info. The info field names\n",
    "# come from https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/\n",
    "ExampleLookupKey = namedtuple('ExampleLookupKey', [\n",
    "    'category',\n",
    "    'instance',\n",
    "    'elevation',\n",
    "    'azimuth',\n",
    "    'lighting',\n",
    "])\n",
    "\n",
    "# Map infos/category to the index of the image in `all_dat`\n",
    "example_lookup = {\n",
    "    ExampleLookupKey(category, *info): i\n",
    "    for i, (info, category) in enumerate(zip(all_info, all_targets))\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Rotating image example\n",
    "\n",
    "I can use the `example_lookup` dictionary to look-up related images. So for example, I can vary nothing but the azimuth."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlMAAABSCAYAAABwglFkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nkikAgEGmAAAGmQIAGAIkVKFhsGUTNQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x10d9076d8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axs = plt.subplots(1, 10, figsize=(10, 1))\n",
    "\n",
    "for ax, azimuth_i in zip(axs, range(10)):\n",
    "    # set everything else arbitrarily\n",
    "    example_key = ExampleLookupKey(\n",
    "        category=4,\n",
    "        instance=8,\n",
    "        elevation=2,\n",
    "        azimuth=2 * azimuth_i,\n",
    "        lighting=2, \n",
    "    )\n",
    "    \n",
    "    example_id = example_lookup[example_key]\n",
    "    # limit to one of the cameras\n",
    "    image = all_dat[example_id][0]\n",
    "    \n",
    "    ax.imshow(image, cmap='gray')\n",
    "    ax.axis('off')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Interactive exploration\n",
    "\n",
    "This one is my favorite, though you'll need this notebook local to view it. I can use the sliders to rotate the toys or view different toys.\n",
    "\n",
    "<img src=\"/assets/2018-05-03-rotate.gif\" width=\"300px\">"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d5405634972442189c0bd15fabbd5b47",
       "version_major": 2,
       "version_minor": 0
      },
      "text/html": [
       "<p>Failed to display Jupyter Widget of type <code>interactive</code>.</p>\n",
       "<p>\n",
       "  If you're reading this message in Jupyter Notebook or JupyterLab, it may mean\n",
       "  that the widgets JavaScript is still loading. If this message persists, it\n",
       "  likely means that the widgets JavaScript library is either not installed or\n",
       "  not enabled. See the <a href=\"https://ipywidgets.readthedocs.io/en/stable/user_install.html\">Jupyter\n",
       "  Widgets Documentation</a> for setup instructions.\n",
       "</p>\n",
       "<p>\n",
       "  If you're reading this message in another notebook frontend (for example, a static\n",
       "  rendering on GitHub or <a href=\"https://nbviewer.jupyter.org/\">NBViewer</a>),\n",
       "  it may mean that your frontend doesn't currently support widgets.\n",
       "</p>\n"
      ],
      "text/plain": [
       "interactive(children=(IntSlider(value=2, description='category', max=4), IntSlider(value=4, description='instance', max=9), IntSlider(value=4, description='elevation', max=8), IntSlider(value=8, description='azimuth', max=17), IntSlider(value=2, description='lighting', max=5), Checkbox(value=False, description='camera_b'), Output()), _dom_classes=('widget-interact',))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<function __main__.draw_image>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from ipywidgets import interact\n",
    "\n",
    "# Hardcode more info from https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/\n",
    "category_idx_to_label = [\n",
    "    'animal',\n",
    "    'human',\n",
    "    'airplane',\n",
    "    'truck',\n",
    "    'car',\n",
    "]\n",
    "\n",
    "elevations_idx_to_label = list(range(30, 72, 5))\n",
    "\n",
    "# Heads up, `info` contains only even values for azimuths\n",
    "# So if I'm not using data from `info`, I need to index into this\n",
    "# by first multiplying by 2\n",
    "azimuth_id_to_label = list(range(0, 360, 10))\n",
    "\n",
    "def draw_image(category, instance, elevation, azimuth, lighting, camera_b):\n",
    "    camera = 1 if camera_b else 0\n",
    "\n",
    "    example_key = ExampleLookupKey(\n",
    "        category=category,\n",
    "        instance=instance,\n",
    "        elevation=elevation,\n",
    "        azimuth=2 * azimuth, # azimuth in info is 2 * the index\n",
    "        lighting=lighting, \n",
    "    )\n",
    "    \n",
    "    example_id = example_lookup[example_key]\n",
    "    \n",
    "    plt.imshow(all_dat[example_id, camera], cmap='gray')\n",
    "    plt.axis('off')\n",
    "    plt.show()\n",
    "\n",
    "    print('category:', category_idx_to_label[category])\n",
    "    print('example:', all_info[example_id][0])\n",
    "    print('elevation:', elevations_idx_to_label[all_info[example_id][1]])\n",
    "    print('azimuth:', azimuth_id_to_label[all_info[example_id][2]])\n",
    "    print('lighting:', all_info[example_id][3])\n",
    "\n",
    "interact(\n",
    "    draw_image,\n",
    "    category=(0, 4),    \n",
    "    instance=(0, 9),\n",
    "    elevation=(0, 8),\n",
    "    azimuth=(0, 17),\n",
    "    lighting=(0, 5),\n",
    "    camera_b=False,    \n",
    ")\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": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}