The 2021 Image Similarity Challenge introduced a dataset to serve as a new benchmark to evaluate recent image copy detection methods. There were 200 participants to the competition. This paper presents a quantitative and qualitative analysis of the top submissions. It appears that the most difficult image transformations involve either severe image crops or hiding into unrelated images, combined with local pixel perturbations. The key algorithmic elements in the winning submissions are: training on strong augmentations, self-supervised learning, score normalization, explicit overlay detection, and global descriptor matching followed by pairwise image comparison.
@article{arxiv.2202.04007,
title = {Results and findings of the 2021 Image Similarity Challenge},
author = {Zoë Papakipos and Giorgos Tolias and Tomas Jenicek and Ed Pizzi and Shuhei Yokoo and Wenhao Wang and Yifan Sun and Weipu Zhang and Yi Yang and Sanjay Addicam and Sergio Manuel Papadakis and Cristian Canton Ferrer and Ondrej Chum and Matthijs Douze},
journal= {arXiv preprint arXiv:2202.04007},
year = {2022}
}