{"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/reco-retrieve-and-co-segment-for-zero-shot-1","title":"ReCo: Retrieve and Co-segment for Zero-shot Transfer","arxiv_id":"2206.07045","date":"2022-06-14","proceeding":null,"authors":["Gyungin Shin","Weidi Xie","Samuel Albanie"],"abstract":"Semantic segmentation has a broad range of applications, but its real-world impact has been significantly limited by the prohibitive annotation costs necessary to enable deployment. Segmentation methods that forgo supervision can side-step these costs, but exhibit the inconvenient requirement to provide labelled examples from the target distribution to assign concept names to predictions. An alternative line of work in language-image pre-training has recently demonstrated the potential to produce models that can both assign names across large vocabularies of concepts and enable zero-shot transfer for classification, but do not demonstrate commensurate segmentation abilities. In this work, we strive to achieve a synthesis of these two approaches that combines their strengths. We leverage the retrieval abilities of one such language-image pre-trained model, CLIP, to dynamically curate training sets from unlabelled images for arbitrary collections of concept names, and leverage the robust correspondences offered by modern image representations to co-segment entities among the resulting collections. The synthetic segment collections are then employed to construct a segmentation model (without requiring pixel labels) whose knowledge of concepts is inherited from the scalable pre-training process of CLIP. We demonstrate that our approach, termed Retrieve and Co-segment (ReCo) performs favourably to unsupervised segmentation approaches while inheriting the convenience of nameable predictions and zero-shot transfer. We also demonstrate ReCo's ability to generate specialist segmenters for extremely rare objects.","url_abs":"https://arxiv.org/abs/2206.07045v1","url_pdf":"https://arxiv.org/pdf/2206.07045v1.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":"reco-retrieve-and-co-segment-for-zero-shot-1","repo_url":"https://github.com/NoelShin/reco","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"reco-retrieve-and-co-segment-for-zero-shot-1","repo_url":"https://github.com/noelshin/namedmask","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"retrieval","task_name":"Retrieval"},{"task_slug":"segmentation","task_name":"Segmentation"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":"unsupervised-semantic-segmentation","task_name":"Unsupervised Semantic Segmentation"},{"task_slug":"unsupervised-semantic-segmentation-with","task_name":"Unsupervised Semantic Segmentation with Language-image Pre-training"}],"methods":[{"method_slug":"clip","method_name":"CLIP"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-4","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"ADE20K","model":"ReCo","rank_in_archive_order":11,"of":13,"metrics":{"Mean IoU (val)":"11.2"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-10","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"COCO-Object","model":"ReCo","rank_in_archive_order":12,"of":12,"metrics":{"mIoU":"15.7"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-9","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"COCO-Stuff-171","model":"ReCo","rank_in_archive_order":11,"of":12,"metrics":{"mIoU":"14.8"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-1","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"COCO-Stuff-27","model":"ReCo+","rank_in_archive_order":1,"of":4,"metrics":{"mIoU":"32.6","pixel accuracy":"54.1"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-1","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"COCO-Stuff-27","model":"ReCo","rank_in_archive_order":3,"of":4,"metrics":{"mIoU":"26.3","pixel accuracy":"46.1"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-3","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"Cityscapes val","model":"ReCo+","rank_in_archive_order":8,"of":12,"metrics":{"mIoU":"24.2","pixel accuracy":"83.7"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-3","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"Cityscapes val","model":"ReCo","rank_in_archive_order":10,"of":12,"metrics":{"mIoU":"19.3","pixel accuracy":"74.6"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-2","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"KITTI-STEP","model":"ReCo+","rank_in_archive_order":1,"of":3,"metrics":{"mIoU":"31.9","pixel accuracy":"75.3"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-2","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"KITTI-STEP","model":"ReCo","rank_in_archive_order":2,"of":3,"metrics":{"mIoU":"29.8","pixel accuracy":"70.6"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-8","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"PASCAL Context-59","model":"ReCo","rank_in_archive_order":12,"of":12,"metrics":{"mIoU":"22.3"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-7","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"PascalVOC-20","model":"ReCo","rank_in_archive_order":10,"of":10,"metrics":{"mIoU":"57.7"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2206.07045","atlas_url":"https://app.syntology.ai/?focus=2206.07045","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}