This data set consists of a set of 10 panorama image sets, together with precomputed local features for each image, and ground truth homographies for each overlapping image pair. The purpose of the data set is to test algorithms for local feature matching in the context of image registration.
For more details, please see the readme document.
The data is available under the MIT license.
If you use the data in a publication, please reference:
Brandt, J. "Transform coding for fast approximate nearest neighbor search in high dimensions", IEEE Conf. on Computer Vision and Pattern Recognition, 2010.
This project is part of the Adobe Open Source Datasets project.