This is W1BS descriptors benchmark from paper WxBS: Wide Baseline Stereo Generalizations .
Dataset format:
data/W1BS/ - directories with subsets. G - geometry, A - appearance, S - sensor, map2photo - map vs. photo
Each directory contains: 1: regerence image dir, 2 - "noised" image dir, h - homography 1to2 dir
each image dir contains several images, e.g. dir (data/W1BS/G/1) =
[arch.keys obama.keys vprice0.keys vprice1.keys vprice2.keys yosemite.keys
arch.png obama.png vprice0.png vprice1.png vprice2.png yosemite.png]
*.png = image, *.keys = text file with affine keypoints in format:
npoints
x y 5.192*s a11 a12 a21 a22
*.bmp - hpatches-style column image with pre-extracted patches
How to get example results (for now, SIFT, BRIEF and ResizeTo11x11 descriptors are available ):
cd data
./download_W1BS_dataset.sh
cd ../code
./do_everything.sh
To add your descriptor to benchmark, please add corresponding script to code/descriptors directory.
The provided file should take two arguments: path to input image input_img.bmp and path to output text file with descriptors.
Output file: one space separated line for one descriptor. Please, see example in code/descriptors/Pixels11.py or code/descriptors/SIFT.py
Please cite us if you use this code:
@InProceedings{Mishkin2015WXBS,
author = {{Mishkin}, D. and {Matas}, J. and {Perdoch}, M. and {Lenc}, K. },
Booktitle = {Proceedings of the British Machine Vision Conference},
Publisher = {BMVA},
title = "{WxBS: Wide Baseline Stereo Generalizations}",
year = 2015,
month = sep,}