Results and findings of the 2021 Image Similarity Challenge
It provides a benchmark for image copy detection, but is incremental as it evaluates existing methods on new data.
The paper analyzed the 2021 Image Similarity Challenge, which introduced a new benchmark dataset for evaluating image copy detection methods, with 200 participants, and found that the most difficult transformations involved severe crops or hidden images with local perturbations, while winning submissions used techniques like strong augmentations and self-supervised learning.
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.