Results and findings of the 2021 Image Similarity Challenge

Zoë Papakipos, Giorgos Tolias, Tomas Jenicek, Ed Pizzi, Shuhei Yokoo, Wenhao Wang, Yifan Sun, Weipu Zhang, Yi Yang, Sanjay Addicam, Sergio Manuel Papadakis, Cristian Canton Ferrer, Ondřej Chum, Matthijs Douze
Proceedings of the NeurIPS 2021 Competitions and Demonstrations Track, PMLR 176:1-12, 2022.

Abstract

The 2021 Image Similarity Challenge introduced a dataset to serve as a benchmark to evaluate 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 overlaying onto 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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v176-papakipos22a, title = {Results and findings of the 2021 Image Similarity Challenge}, author = {Papakipos, Zo\"e and Tolias, Giorgos and Jenicek, Tomas and Pizzi, Ed and Yokoo, Shuhei and Wang, Wenhao and Sun, Yifan and Zhang, Weipu and Yang, Yi and Addicam, Sanjay and Papadakis, Sergio Manuel and Ferrer, Cristian Canton and Chum, Ond{\v{r}}ej and Douze, Matthijs}, booktitle = {Proceedings of the NeurIPS 2021 Competitions and Demonstrations Track}, pages = {1--12}, year = {2022}, editor = {Kiela, Douwe and Ciccone, Marco and Caputo, Barbara}, volume = {176}, series = {Proceedings of Machine Learning Research}, month = {06--14 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v176/papakipos22a/papakipos22a.pdf}, url = {https://proceedings.mlr.press/v176/papakipos22a.html}, abstract = {The 2021 Image Similarity Challenge introduced a dataset to serve as a benchmark to evaluate 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 overlaying onto 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.} }
Endnote
%0 Conference Paper %T Results and findings of the 2021 Image Similarity Challenge %A Zoë Papakipos %A Giorgos Tolias %A Tomas Jenicek %A Ed Pizzi %A Shuhei Yokoo %A Wenhao Wang %A Yifan Sun %A Weipu Zhang %A Yi Yang %A Sanjay Addicam %A Sergio Manuel Papadakis %A Cristian Canton Ferrer %A Ondřej Chum %A Matthijs Douze %B Proceedings of the NeurIPS 2021 Competitions and Demonstrations Track %C Proceedings of Machine Learning Research %D 2022 %E Douwe Kiela %E Marco Ciccone %E Barbara Caputo %F pmlr-v176-papakipos22a %I PMLR %P 1--12 %U https://proceedings.mlr.press/v176/papakipos22a.html %V 176 %X The 2021 Image Similarity Challenge introduced a dataset to serve as a benchmark to evaluate 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 overlaying onto 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.
APA
Papakipos, Z., Tolias, G., Jenicek, T., Pizzi, E., Yokoo, S., Wang, W., Sun, Y., Zhang, W., Yang, Y., Addicam, S., Papadakis, S.M., Ferrer, C.C., Chum, O. & Douze, M.. (2022). Results and findings of the 2021 Image Similarity Challenge. Proceedings of the NeurIPS 2021 Competitions and Demonstrations Track, in Proceedings of Machine Learning Research 176:1-12 Available from https://proceedings.mlr.press/v176/papakipos22a.html.

Related Material