Binary Descriptor Evaluation

This work evaluates the performance of binary descriptors and their pairings with various detectors.
ECCV 2012 PDF
ECCV 2012 Poster
ECCV 2012 BibTeX
Datasets
Download Oxford Datasets
[1]
Download Strecha Datasets
[2]
Download Photo-Collection and Additional Datasets (~8 MB)
[3]
Download All Data (images, camera parameters, keypoints) (~85 MB)
Strecha Dataset Notes
Each image is resized so that its width is 1024.
For the fountain-P11 dataset, 0005.png was taken to be the first image, with 0004.png following, and then sequentially down to 0000.png (6 images total).
Example Dataset Images
- Bikes (Blur) [1]
- Trees (Blur) [1]
- UBC (JPEG Compression) [1]
- Leuven (Exposure) [1]
- Graffiti (Viewpoint) [1]
- Wall (Viewpoint) [1]
- Bark (Zoom, Rotation) [1]
- Boat (Zoom, Rotation) [1]
- Fountain-P11 (Viewpoint) [2]
- Herz-Jesu-P8 (Viewpoint) [2]
- Reichstag (Viewpoint, Zoom, Illumination) [3]
- Berliner-Dom (Viewpoint, Zoom, Illumination) [3]
- Day-Night (Illumination) [3]
- Semper (Rotation) [3]
- Venice (Zoom) [3]
- Rome (Rotation) [3]
- Ceiling (Rotation) [3]
References
[1] K. Mikolajczyk, T. Tuytelaars, C. Schmid, A. Zisserman, J. Matas, F. Schaffalitzky, T. Kadir, and L.V. Gool, "A comparison of affine region detectors," IJCV, 65:43-72, November 2005.
[2] C. Strecha, W. von Hansen, L. Van Gool, P. Fua, U. Thoennessen, "On Benchmarking Camera Calibration and Multi-View Stereo for High Resolution Imagery," CVPR, 2008.
[3] J. Heinly, E. Dunn, J.M. Frahm, "Comparative Evaluation of Binary Features," ECCV, 2012.