{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/exploring-the-limits-of-deep-image-clustering","title":"Exploring the Limits of Deep Image Clustering using Pretrained Models","arxiv_id":"2303.17896","date":"2023-03-31","proceeding":null,"authors":["Nikolas Adaloglou","Felix Michels","Hamza Kalisch","Markus Kollmann"],"abstract":"We present a general methodology that learns to classify images without labels by leveraging pretrained feature extractors. Our approach involves self-distillation training of clustering heads based on the fact that nearest neighbours in the pretrained feature space are likely to share the same label. We propose a novel objective that learns associations between image features by introducing a variant of pointwise mutual information together with instance weighting. We demonstrate that the proposed objective is able to attenuate the effect of false positive pairs while efficiently exploiting the structure in the pretrained feature space. As a result, we improve the clustering accuracy over $k$-means on $17$ different pretrained models by $6.1$\\% and $12.2$\\% on ImageNet and CIFAR100, respectively. Finally, using self-supervised vision transformers, we achieve a clustering accuracy of $61.6$\\% on ImageNet. The code is available at https://github.com/HHU-MMBS/TEMI-official-BMVC2023.","url_abs":"https://arxiv.org/abs/2303.17896v2","url_pdf":"https://arxiv.org/pdf/2303.17896v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"exploring-the-limits-of-deep-image-clustering","repo_url":"https://github.com/HHU-MMBS/TEMI-official-BMVC2023","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"clustering","task_name":"Clustering"},{"task_slug":"image-clustering","task_name":"Image Clustering"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-clustering-on-cifar-10","task":"Image Clustering","dataset":"CIFAR-10","model":"TEMI CLIP ViT-L (openai)","rank_in_archive_order":4,"of":40,"metrics":{"ARI":"0.932","Accuracy":"0.969","Backbone":"ViT-L","NMI":"0.926","Train set":"Train"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-cifar-10","task":"Image Clustering","dataset":"CIFAR-10","model":"TEMI DINO ViT-B","rank_in_archive_order":40,"of":40,"metrics":{"ARI":"0.885","Accuracy":"0.94.5","Backbone":"ViT-B","NMI":"0.886","Train set":"Train"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-cifar-100","task":"Image Clustering","dataset":"CIFAR-100","model":"TEMI CLIP ViT-L (openai)","rank_in_archive_order":4,"of":30,"metrics":{"ARI":"0.612","Accuracy":"0.737","NMI":"0.799","Train Set":"Train"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-cifar-100","task":"Image Clustering","dataset":"CIFAR-100","model":"TEMI DINO ViT-B","rank_in_archive_order":5,"of":30,"metrics":{"ARI":"0.533","Accuracy":"0.671","NMI":"0.769","Train Set":"Train"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-imagenet","task":"Image Clustering","dataset":"ImageNet","model":"TEMI MSN (ViT-L)","rank_in_archive_order":5,"of":12,"metrics":{"ARI":"48.4","Accuracy":"61.6","NMI":"82.5"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-imagenet","task":"Image Clustering","dataset":"ImageNet","model":"TEMI DINO (ViT-B)","rank_in_archive_order":7,"of":12,"metrics":{"ARI":"45.9","Accuracy":"58.0","NMI":"81.4"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-imagenet-100","task":"Image Clustering","dataset":"ImageNet-100 (TEMI Split)","model":"TEMI CLIP ViT-L (openai)","rank_in_archive_order":1,"of":5,"metrics":{"ACCURACY":"0.8343","ARI":"0.7581","NMI":"0.9006"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-imagenet-100","task":"Image Clustering","dataset":"ImageNet-100 (TEMI Split)","model":"TEMI MSN ViT-L","rank_in_archive_order":2,"of":5,"metrics":{"ACCURACY":"0.8286","ARI":"0.7408","NMI":"0.8853"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-imagenet-100","task":"Image Clustering","dataset":"ImageNet-100 (TEMI Split)","model":"TEMI DINO ViT-B","rank_in_archive_order":3,"of":5,"metrics":{"ACCURACY":"0.7505","ARI":"0.6545","NMI":"0.8565"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-imagenet-200","task":"Image Clustering","dataset":"ImageNet-200","model":"TEMI CLIP ViT-L (openai)","rank_in_archive_order":1,"of":5,"metrics":{"\t ACCURACY":"0.7776","ARI":"0.6941","NMI":"0.8839"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-imagenet-200","task":"Image Clustering","dataset":"ImageNet-200","model":"TEMI MSN ViT-L","rank_in_archive_order":2,"of":5,"metrics":{"\t ACCURACY":"0.77.96","ARI":"0.667","NMI":"0.8665"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-imagenet-200","task":"Image Clustering","dataset":"ImageNet-200","model":"TEMI DINO ViT-B","rank_in_archive_order":3,"of":5,"metrics":{"\t ACCURACY":"0.7312","ARI":"0.6231","NMI":"0.852"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-imagenet-50-1","task":"Image Clustering","dataset":"ImageNet-50 (TEMI Split)","model":"TEMI CLIP ViT-L (openai)","rank_in_archive_order":1,"of":5,"metrics":{"ACCURACY":"0.8827","ARI":"0.8272","NMI":"0.9232"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-imagenet-50-1","task":"Image Clustering","dataset":"ImageNet-50 (TEMI Split)","model":"TEMI MSN ViT-L","rank_in_archive_order":2,"of":5,"metrics":{"ACCURACY":"0.8487","ARI":"0.7646","NMI":"0.8814"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-imagenet-50-1","task":"Image Clustering","dataset":"ImageNet-50 (TEMI Split)","model":"TEMI DINO ViT-B","rank_in_archive_order":3,"of":5,"metrics":{"ACCURACY":"0.801","ARI":"0.7093","NMI":"0.8610"},"uses_additional_data":true},{"leaderboard":"/sota/image-clustering-on-stl-10","task":"Image Clustering","dataset":"STL-10","model":"TEMI DINO ViT-B","rank_in_archive_order":2,"of":29,"metrics":{"ARI":"0.968","Accuracy":"0.985","Backbone":"ViT-B","NMI":"0.965","Train Split":"Train"},"uses_additional_data":true}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2303.17896","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2303.17896"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/HHU-MMBS/TEMI-official-BMVC2023","reach":null}],"summary":{"ran_fixture":1,"ran_draft_wrong":3},"by_repo_kind":{"official":{"samples":4,"ran":4,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"a4e72aa97d72ef5e","entry":"compute_embedding","repo":"HHU-MMBS/TEMI-official-BMVC2023","repo_kind":"official","path":"gen_embeds.py","file_url":"https://github.com/HHU-MMBS/TEMI-official-BMVC2023/blob/HEAD/gen_embeds.py","link_basis":"first_harvest_node","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"a4e72aa97d72ef5e"}},{"code_sha256_prefix":"709a9a2f23ee145d","entry":"compute_neighbors","repo":"HHU-MMBS/TEMI-official-BMVC2023","repo_kind":"official","path":"gen_embeds.py","file_url":"https://github.com/HHU-MMBS/TEMI-official-BMVC2023/blob/HEAD/gen_embeds.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"MISDECLARED","metamorphic_tier":"invariant","behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"709a9a2f23ee145d"}},{"code_sha256_prefix":"79b303c122ae61f6","entry":"default_out_dir","repo":"HHU-MMBS/TEMI-official-BMVC2023","repo_kind":"official","path":"train_main.py","file_url":"https://github.com/HHU-MMBS/TEMI-official-BMVC2023/blob/HEAD/train_main.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"79b303c122ae61f6"}},{"code_sha256_prefix":"5914e91451bf1c5e","entry":"get_outpath","repo":"HHU-MMBS/TEMI-official-BMVC2023","repo_kind":"official","path":"gen_embeds.py","file_url":"https://github.com/HHU-MMBS/TEMI-official-BMVC2023/blob/HEAD/gen_embeds.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"5914e91451bf1c5e"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}