A Vote-and-Verify Strategy for Fast Spatial Verification in Image Retrieval

This is the official repository for the publication:

@inproceedings{schoenberger2016vote,
    author = {Sch\"{o}nberger, Johannes Lutz and Price, True and Sattler, Torsten and Frahm, Jan-Michael and Pollefeys, Marc},
    title = {A Vote-and-Verify Strategy for Fast Spatial Verification in Image Retrieval},
    booktitle={Asian Conference on Computer Vision (ACCV)},
    year={2016},
}

World5k Dataset

The URLs for all images in our dataset can be found in the image_urls folder.

Vote-and-Verify Code

An implementation of both the vocabulary tree with Hamming embedding as well as our proposed Vote-and-Verify method can be found in COLMAP (https://github.com/colmap/colmap). COLMAP is a Structure-from-Motion and Multi- View Stereo library. COLMAP implements a fully functional image retrieval system (in the src/retrieval/* folder), that can be used with the executables:

• src/exe/vocab_tree_builder: to build a custom vocabulary tree from image features

• src/exe/vocab_tree_retriever: to perform image retrieval using a pre-built vocabulary tree

• src/exe/vocab_tree_matcher: to match images using the vocabulary tree

The number of images to re-rank during spatial verification can be specified using the num_verifications option. Please refer to the code and the documentation of COLMAP for more details and fine-grain control of the parameters (https://colmap.github.io/).