Initial commit

2_finetuning-alexnet-wikiart-style/download_wikiart.py

@@ -0,0 +1,86 @@
+#!/usr/bin/env python
+"""
+Download WikiArt images and write
+Caffe ImageData layer files
+
+This script has been forked from
+https://github.com/BVLC/caffe/blob/master/examples/finetune_flickr_style/assemble_data.py
+"""
+import os
+import urllib
+import argparse
+import numpy as np
+import pandas as pd
+from skimage import io
+import multiprocessing
+
+root_dirname = os.path.abspath(os.path.dirname(__file__))
+training_dirname = os.path.join(root_dirname, 'data/wikiart')
+
+def download_image(args_tuple):
+    "For use with multiprocessing map. Returns filename on fail."
+    try:
+        url, filename = args_tuple
+        if not os.path.exists(filename):
+            urllib.urlretrieve(url, filename)
+        test_read_image = io.imread(filename)
+        return True
+    except KeyboardInterrupt:
+        raise Exception()  # multiprocessing doesn't catch keyboard exceptions
+    except:
+        return False
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(
+        description='Download a subset of the WikiArt style dataset to a directory.')
+    parser.add_argument(
+        '-s', '--seed', type=int, default=0,
+        help="random seed")
+    parser.add_argument(
+        '-i', '--images', type=int, default=-1,
+        help="number of images to use (-1 for all [default])",
+    )
+    parser.add_argument(
+        '-w', '--workers', type=int, default=-1,
+        help="num workers used to download images. -x uses (all - x) cores [-1 default]."
+    )
+
+    args = parser.parse_args()
+    np.random.seed(args.seed)
+
+    # Read data, shuffle order, and subsample.
+    csv_filename = os.path.join(root_dirname, 'wikiart.csv.gz')
+    df = pd.read_csv(csv_filename, index_col=0, compression='gzip')
+    df = df.iloc[np.random.permutation(df.shape[0])]
+    if args.images > 0 and args.images < df.shape[0]:
+        df = df.iloc[:args.images]
+
+    # Make directory for images and get local filenames.
+    if training_dirname is None:
+        training_dirname = os.path.join(root_dirname, 'data/wikiart')
+    images_dirname = os.path.join(training_dirname, 'images')
+    if not os.path.exists(images_dirname):
+        os.makedirs(images_dirname)
+    df['image_filename'] = [
+        os.path.join(images_dirname, _ + '.jpg') for _ in df.index.values
+    ]
+
+    # Download images.
+    num_workers = args.workers
+    if num_workers <= 0:
+        num_workers = multiprocessing.cpu_count() + num_workers
+    print('Downloading {} images with {} workers...'.format(
+        df.shape[0], num_workers))
+    pool = multiprocessing.Pool(processes=num_workers)
+    map_args = zip(df['image_url'], df['image_filename'])
+    results = pool.map(download_image, map_args)
+
+    # Only keep rows with valid images, and write out training file lists.
+    df = df[results]
+    for split in ['train', 'test']:
+        split_df = df[df['_split'] == split]
+        filename = os.path.join(training_dirname, '{}.txt'.format(split))
+        split_df[['image_filename', 'label']].to_csv(
+            filename, sep=' ', header=None, index=None)
+    print('Writing train/val for {} successfully downloaded images.'.format(
+        df.shape[0]))

2_finetuning-alexnet-wikiart-style/styles.csv

@@ -0,0 +1,11 @@
+label,style
+0,Art Nouveau (Modern)
+1,Baroque
+2,Expressionism
+3,Impressionism
+4,Neoclassicism
+5,Post-Impressionism
+6,Realism
+7,Romanticism
+8,Surrealism
+9,Symbolism

3_visualizing-breaking-convnets/breaking-convnets.ipynb

@@ -0,0 +1,163 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "# Setup\n",
+    "import os\n",
+    "import numpy as np\n",
+    "import matplotlib.pyplot as plt\n",
+    "import matplotlib.cm as cm\n",
+    "%matplotlib inline\n",
+    "\n",
+    "# Make sure that caffe is on the python path\n",
+    "caffe_root = ''\n",
+    "import sys\n",
+    "sys.path.insert(0, caffe_root + 'python')\n",
+    "\n",
+    "import caffe\n",
+    "\n",
+    "plt.rcParams['figure.figsize'] = (10, 10)\n",
+    "plt.rcParams['image.interpolation'] = 'nearest'\n",
+    "plt.rcParams['image.cmap'] = 'gray'\n",
+    "\n",
+    "model_prototxt = 'deploy.prototxt'\n",
+    "pretrained_model = os.path.join(caffe_root,'models/bvlc_alexnet/bvlc_alexnet.caffemodel')\n",
+    "\n",
+    "caffe.set_mode_cpu()\n",
+    "net = caffe.Classifier(model_prototxt, pretrained_model,\n",
+    "                       mean=np.load(os.path.join(caffe_root,'python/caffe/imagenet/ilsvrc_2012_mean.npy')).mean(1).mean(1),\n",
+    "                       channel_swap=(2,1,0),\n",
+    "                       raw_scale=255,\n",
+    "                       image_dims=(256, 256))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We start with the image of a cat and desired label as goose. We update input image for `n_iterations` iterations till it maximises class score for goose."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "image_file = 'cat_2.jpg'\n",
+    "input_image = caffe.io.load_image(image_file)\n",
+    "\n",
+    "n_iterations = 40\n",
+    "desired_label = 99 # goose\n",
+    "label = np.zeros((1,1,1,1000))\n",
+    "label[0,0,0,desired_label] = 1;\n",
+    "step_size = 1000\n",
+    "reg = 0.002\n",
+    "\n",
+    "input_image = net.transformer.preprocess('data',input_image)\n",
+    "fooling_image = input_image[:]\n",
+    "zero_image = np.zeros(fooling_image.shape)\n",
+    "\n",
+    "for i in range(n_iterations):\n",
+    "    net.blobs['data'].data[...] = fooling_image\n",
+    "    \n",
+    "    # Perform forward pass\n",
+    "    # TODO\n",
+    "    # \n",
+    "    # END OF YOUR CODE\n",
+    "    \n",
+    "    # Perform backward pass for the desired class\n",
+    "    # TODO\n",
+    "    # \n",
+    "    # END OF YOUR CODE\n",
+    "    \n",
+    "    # Compute gradient and incremental update\n",
+    "    # Store update value in di\n",
+    "    # TODO\n",
+    "    # \n",
+    "    # di = \n",
+    "    # END OF YOUR CODE\n",
+    "    \n",
+    "    fooling_image += di\n",
+    "    zero_image += di\n",
+    "    \n",
+    "plt.subplot(1,2,1)\n",
+    "plt.imshow(net.transformer.deprocess('data', zero_image))\n",
+    "plt.subplot(1,2,2)\n",
+    "plt.imshow(net.transformer.deprocess('data', fooling_image))\n",
+    "\n",
+    "# Save the image\n",
+    "plt.imsave('cat_fooled.jpg',net.transformer.deprocess('data', fooling_image))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Here we check our prediction on the modified image to make sure it is able to 'fool' the ConvNet."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "['n01855672 goose' 'n01806567 quail' 'n01847000 drake'\n",
+      " 'n02018207 American coot, marsh hen, mud hen, water hen, Fulica americana'\n",
+      " 'n01592084 chickadee']\n"
+     ]
+    }
+   ],
+   "source": [
+    "image_file = 'cat_fooled.jpg'\n",
+    "net.blobs['data'].data[...] = net.transformer.preprocess('data',caffe.io.load_image(image_file))\n",
+    "prediction = net.forward()\n",
+    "# load labels\n",
+    "imagenet_labels_filename = os.path.join(caffe_root,'data/ilsvrc12/synset_words.txt')\n",
+    "try:\n",
+    "    labels = np.loadtxt(imagenet_labels_filename, str, delimiter='\\t')\n",
+    "except:\n",
+    "    !../data/ilsvrc12/get_ilsvrc_aux.sh\n",
+    "    labels = np.loadtxt(imagenet_labels_filename, str, delimiter='\\t')\n",
+    "# sort top k predictions from softmax output\n",
+    "top_k = prediction['fc8'][0].flatten().argsort()[-1:-6:-1]\n",
+    "print labels[top_k]"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 2",
+   "language": "python",
+   "name": "python2"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 2
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython2",
+   "version": "2.7.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 0
+}