{"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/let-go-of-your-labels-with-unsupervised-1","title":"Let Go of Your Labels with Unsupervised Transfer","arxiv_id":"2406.07236","date":"2024-06-11","proceeding":"International Conference on Machine Learning 2024 6","authors":["Artyom Gadetsky","Yulun Jiang","Maria Brbic"],"abstract":"Foundation vision-language models have enabled remarkable zero-shot transferability of the pre-trained representations to a wide range of downstream tasks. However, to solve a new task, zero-shot transfer still necessitates human guidance to define visual categories that appear in the data. Here, we show that fully unsupervised transfer emerges when searching for the labeling of a dataset that induces maximal margin classifiers in representation spaces of different foundation models. We present TURTLE, a fully unsupervised method that effectively employs this guiding principle to uncover the underlying labeling of a downstream dataset without any supervision and task-specific representation learning. We evaluate TURTLE on a diverse benchmark suite of 26 datasets and show that it achieves new state-of-the-art unsupervised performance. Furthermore, TURTLE, although being fully unsupervised, outperforms zero-shot transfer baselines on a wide range of datasets. In particular, TURTLE matches the average performance of CLIP zero-shot on 26 datasets by employing the same representation space, spanning a wide range of architectures and model sizes. By guiding the search for the underlying labeling using the representation spaces of two foundation models, TURTLE surpasses zero-shot transfer and unsupervised prompt tuning baselines, demonstrating the surprising power and effectiveness of unsupervised transfer.","url_abs":"https://arxiv.org/abs/2406.07236v1","url_pdf":"https://arxiv.org/pdf/2406.07236v1.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":"let-go-of-your-labels-with-unsupervised-1","repo_url":"https://github.com/mlbio-epfl/turtle","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"image-clustering","task_name":"Image Clustering"},{"task_slug":"unsupervised-image-classification","task_name":"Unsupervised Image Classification"}],"methods":[{"method_slug":"clip","method_name":"CLIP"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-clustering-on-birdsnap","task":"Image Clustering","dataset":"Birdsnap","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"68.1"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-cifar-10","task":"Image Clustering","dataset":"CIFAR-10","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":40,"metrics":{"ARI":"0.989","Accuracy":"0.995","NMI":"0.985"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-cifar-100","task":"Image Clustering","dataset":"CIFAR-100","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":30,"metrics":{"ARI":"0.834","Accuracy":"0.898","NMI":"0.915"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-clevr-counts","task":"Image Clustering","dataset":"CLEVR Counts","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"24.0"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-caltech-101","task":"Image Clustering","dataset":"Caltech-101","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"89.8"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-country211","task":"Image Clustering","dataset":"Country211","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"11.1"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-dtd","task":"Image Clustering","dataset":"DTD","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":2,"of":2,"metrics":{"Accuracy":"57.3"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-eurosat","task":"Image Clustering","dataset":"EuroSAT","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"96.6"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-fer2013","task":"Image Clustering","dataset":"FER2013","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"36.2"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-fgvc-aircraft","task":"Image Clustering","dataset":"FGVC Aircraft","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"36.5"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-flowers-102","task":"Image Clustering","dataset":"Flowers-102","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"99.6"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-food-101","task":"Image Clustering","dataset":"Food-101","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"92.2"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-gtsrb","task":"Image Clustering","dataset":"GTSRB","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"48.4"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-hateful-memes","task":"Image Clustering","dataset":"Hateful Memes","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"54.2"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-imagenet","task":"Image Clustering","dataset":"ImageNet","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":12,"metrics":{"ARI":"62.5","Accuracy":"72.9","NMI":"88.2"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-kitti","task":"Image Clustering","dataset":"KITTI","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"39.4"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-kinetics-700","task":"Image Clustering","dataset":"Kinetics-700","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"43.0"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-mnist","task":"Image Clustering","dataset":"MNIST","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":3,"metrics":{"Accuracy":"97.8"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-oxford-iiit-pets","task":"Image Clustering","dataset":"Oxford-IIIT Pets","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"92.3"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-pcam","task":"Image Clustering","dataset":"PCam","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"52.0"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-resisc45","task":"Image Clustering","dataset":"RESISC45","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"89.6"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-rendered-sst2","task":"Image Clustering","dataset":"Rendered SST2","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"51.6"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-stl-10","task":"Image Clustering","dataset":"STL-10","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":29,"metrics":{"ARI":"0.994","Accuracy":"0.997","NMI":"0.993"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-sun397","task":"Image Clustering","dataset":"SUN397","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"67.9"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-stanford-cars","task":"Image Clustering","dataset":"Stanford Cars","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":5,"metrics":{"Accuracy":"0.646"},"uses_additional_data":false},{"leaderboard":"/sota/image-clustering-on-ucf101","task":"Image Clustering","dataset":"UCF101","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":2,"metrics":{"Accuracy":"82.3"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-image-classification-on-cifar-10","task":"Unsupervised Image Classification","dataset":"CIFAR-10","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":9,"metrics":{"Accuracy":"99.5"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-image-classification-on-imagenet","task":"Unsupervised Image Classification","dataset":"ImageNet","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":9,"metrics":{"ARI":"62.5","Accuracy (%)":"72.9"},"uses_additional_data":true},{"leaderboard":"/sota/unsupervised-image-classification-on-mnist","task":"Unsupervised Image Classification","dataset":"MNIST","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":3,"of":10,"metrics":{"Accuracy":"97.8"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-image-classification-on-stl-10","task":"Unsupervised Image Classification","dataset":"STL-10","model":"TURTLE (CLIP + DINOv2)","rank_in_archive_order":1,"of":9,"metrics":{"Accuracy":"99.7"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2406.07236","atlas_url":"https://app.syntology.ai/?focus=2406.07236","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2406.07236"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-25T09:33:49+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/mlbio-epfl/turtle","reach":null}],"summary":{"ran_draft_wrong":2,"ran_honours":1,"unverified":1},"by_repo_kind":{"official":{"samples":4,"ran":3,"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":4,"samples":[{"code_sha256_prefix":"806241343c4b7e7b","entry":"_parse_args","repo":"mlbio-epfl/turtle","repo_kind":"official","path":"run_turtle.py","file_url":"https://github.com/mlbio-epfl/turtle/blob/HEAD/run_turtle.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":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"806241343c4b7e7b"}},{"code_sha256_prefix":"89923b3abde90046","entry":"get_cluster_acc","repo":"mlbio-epfl/turtle","repo_kind":"official","path":"run_turtle.py","file_url":"https://github.com/mlbio-epfl/turtle/blob/HEAD/run_turtle.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"89923b3abde90046"}},{"code_sha256_prefix":"b300ef9fae8f3a37","entry":"get_features","repo":"mlbio-epfl/turtle","repo_kind":"official","path":"precompute_representations.py","file_url":"https://github.com/mlbio-epfl/turtle/blob/HEAD/precompute_representations.py","link_basis":"first_harvest_node","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":"well_formed","behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"b300ef9fae8f3a37"}},{"code_sha256_prefix":"b22aeb3079826cd8","entry":"run","repo":"mlbio-epfl/turtle","repo_kind":"official","path":"run_turtle.py","file_url":"https://github.com/mlbio-epfl/turtle/blob/HEAD/run_turtle.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"b22aeb3079826cd8"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}