User Friendly README, Batch Extraction with Python Multiprocessing, RANSAC Visualization with Python Notebook

.gitignore

@@ -1 +1,4 @@
 hesaff
+opencv
+.ipynb_checkpoints
+*.swp

README → README.md

@@ -1,4 +1,4 @@
-DESCRIPTION
+# DESCRIPTION
 
 This is an implementation of Hessian-Affine detector. 
 
@@ -12,7 +12,37 @@ neighbourhood is normalized to a fixed size patch and SIFT
 descriptor(Lowe 1999, Lowe 2004) computed.
 
 
-IMPLEMENTATION
+# IMPLEMENTATION
+
+## Installing OpenCV
+```
+git clone https://github.com/opencv/opencv.git
+cd opencv
+git checkout 2154e62 # Checkout for 2.3.1 version commit point
+mkdir build
+cd build
+# This old OpenCV may not support your recent GPU. Just disable it.
+cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D WITH_CUDA=OFF ..
+make -j8
+sudo make install
+```
+
+## Build
+
+```
+cd hesaff
+export PKG_CONFIG_PATH=/usr/local/lib/pkgconfig
+make
+```
+
+## Run
+
+```
+# Assume that libopencv_core.so.2.3 in in /usr/local/lib
+export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib
+# Usage: hesaff image_name.ppm
+hesaff oxford_001.ppm
+```
 
 Implementation depends on OpenCV (2.3.1+). Although, the code is
 original, the affine iteration and normalization was derived from the
@@ -24,7 +54,7 @@ use of SIFT code, files siftdesc.cpp/siftdesc.h, in commercial
 applications (see LICENSE.SIFT for details)
 
 
-OUTPUT
+# OUTPUT
 
 The built binary rewrites output file: <input_image_name>.hesaff.sift
 
@@ -44,7 +74,7 @@ u1 v1 a1 b1 c1 d1(1) d1(2) d1(3) ... d1(N)
 um vm am bm cm dm(1) dm(2) dm(3) ... dm(N)
 
 
-PROPER USE
+# PROPER USE
 
 If you use this code, please refer to
 
@@ -53,4 +83,4 @@ Local Geometry for Large Scale Object Retrieval. In proceedings of
 CVPR09. June 2009.
 
 TBD: A reference to technical report describing the details and some
-retrieval results will be placed here.
+retrieval results will be placed here.

util/Batch SIFT Extractor.ipynb

@@ -0,0 +1,123 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Batch SIFT Extractor\n",
+    "\n",
+    "## Requirement\n",
+    "\n",
+    "* Python3\n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "print(os.getcwd())\n",
+    "# DATA_DIR_OXF5K = \"../../../data/oxf5k_images\" # 5062 images\n",
+    "# OUTPUT_DIR = \"../../../data/oxf5k_hesaff/\"\n",
+    "\n",
+    "# DATA_DIR_OXF100K = \"../data/oxford/oxc1_100k\" # The Flickr 100k dataset consists of 100071 images collected from Flickr by searching for popular Flickr tags. \n",
+    "# 5602 + 100071 = 105,673\n",
+    "\n",
+    "DATA_DIR_PAR6K = \"../../../data/paris6k_images\" # 6392 images\n",
+    "OUTPUT_DIR = \"../../../data/paris6k_hesaff/\"\n",
+    "\n",
+    "if not os.path.exists(OUTPUT_DIR):\n",
+    "    os.mkdir(OUTPUT_DIR)\n",
+    "\n",
+    "def get_images(data_dir):\n",
+    "    image_paths = []\n",
+    "    for parent_dir, _, files in os.walk(data_dir):\n",
+    "        # print(parent_dir)\n",
+    "        for f in files:\n",
+    "            # print(f)\n",
+    "            if f[-4:] == \".jpg\":\n",
+    "                image_paths.append(os.path.abspath(os.path.join(parent_dir, f)))\n",
+    "            else:\n",
+    "                print(\"skip:\", f)\n",
+    "    # filenames = os.listdir(data_dir)   \n",
+    "    # image_paths = list(map(lambda x: os.path.join(data_dir, x), filenames))\n",
+    "    return image_paths\n",
+    "\n",
+    "image_list = get_images(DATA_DIR_PAR6K)\n",
+    "# image_list = get_images(DATA_DIR_OXF5K)\n",
+    "# image_list = get_images(DATA_DIR_OXF5K) + get_images(DATA_DIR_OXF100K)\n",
+    "image_list.sort()\n",
+    "\n",
+    "for i in image_list[:5]:\n",
+    "    print(i)\n",
+    "print(len(image_list))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import subprocess\n",
+    "from subprocess import CalledProcessError\n",
+    "import numpy as np\n",
+    "import shutil\n",
+    "# from tqdm import tqdm_notebook as tqdm\n",
+    "from tqdm import tqdm\n",
+    "\n",
+    "from multiprocessing import Pool\n",
+    "\n",
+    "def extract_sift(image_path):\n",
+    "    try:\n",
+    "        output = subprocess.check_output([\"../hesaff\", \"{}\".format(image_path)])\n",
+    "        # print('stdout of hesaff:', output)\n",
+    "    except CalledProcessError as e:       \n",
+    "        print(e)\n",
+    "        print(\"output: {}\".format(e.output.decode('utf-8')))   \n",
+    "    \n",
+    "\n",
+    "\n",
+    "def parallel_task(image_path):\n",
+    "    extract_sift(image_path)\n",
+    "    sift_file_path = image_path + \".hesaff.sift\"\n",
+    "    sift_file_name = os.path.basename(sift_file_path)\n",
+    "    shutil.move(sift_file_path, os.path.join(OUTPUT_DIR, sift_file_name))\n",
+    "\n",
+    "if __name__ == '__main__':\n",
+    "    pool = Pool(processes=40)\n",
+    "    for result in tqdm(pool.imap_unordered(parallel_task, image_list), total=len(image_list)):\n",
+    "        pass\n",
+    "    # features.append(result)\n",
+    "    \n",
+    "# Timing history: 100k images took about 1h 20m for feature extraction. \n",
+    "\n",
+    "# hesaff speed: 31 images / sec, 32 ms / image"
+   ]
+  }
+ ],
+ "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": "3.6.4"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}