Visualizing ConvNet Activation Maps (PyTorch).ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from matplotlib import gridspec\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import torch\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def visualize_img_activations(img, activations, cmap=\"gray\", \n",
    "                              prefix=\"\", luminance_scale=(None, None),\n",
    "                              fontsize=24):\n",
    "    \n",
    "    h, w = img.shape\n",
    "    k, ah, aw = activations.shape\n",
    "    vmin, vmax = luminance_scale\n",
    "    \n",
    "    # Activations subplot: nrow x ncol\n",
    "    nrows = int(np.ceil(np.sqrt(k)))\n",
    "    ncols = int(np.ceil(1. * k / nrows))\n",
    "    \n",
    "    # Input image span\n",
    "    img_rows = nrows if nrows < 16 else nrows // 2\n",
    "    img_cols = img_rows\n",
    "    \n",
    "    # Main grid dimensions\n",
    "    grid_rows, grid_cols = nrows, ncols + img_cols\n",
    "    gridspec.GridSpec(grid_rows, grid_cols)\n",
    "    \n",
    "    # Plot image\n",
    "    plt.subplot2grid((grid_rows, grid_cols), (0,0), colspan=img_cols, rowspan=img_rows)\n",
    "    \n",
    "    if vmin:\n",
    "        plt.imshow(img, cmap, interpolation=None, vmin=vmin, vmax=vmax)\n",
    "    else:\n",
    "        plt.imshow(img, cmap, interpolation=None)\n",
    "        \n",
    "    plt.xticks([]), plt.yticks([])\n",
    "    plt.ylabel(\"Input\", fontsize=fontsize)\n",
    "    \n",
    "    # Plot activations\n",
    "    for r in range(nrows):\n",
    "        for c in range(ncols):\n",
    "            \n",
    "            a_idx = r * ncols + c\n",
    "            if a_idx < len(activations):\n",
    "                plt.subplot2grid((grid_rows, grid_cols), (r, img_cols+c), colspan=1, rowspan=1)\n",
    "                if r == 0 and c == int(np.floor(ncols/2)):\n",
    "                    plt.title(\"{} {:s}\".format(prefix, \"Activations\"), fontsize=fontsize)\n",
    "                if vmin:\n",
    "                    plt.imshow(activations[a_idx], cmap, interpolation=None, vmin=vmin, vmax=vmax)\n",
    "                else:\n",
    "                    plt.imshow(activations[a_idx], cmap, interpolation=None)\n",
    "                    \n",
    "            plt.xticks([]), plt.yticks([])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "irl_python3",
   "language": "python",
   "name": "irl_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
}