{"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/ttd-text-tag-self-distillation-enhancing","title":"TTD: Text-Tag Self-Distillation Enhancing Image-Text Alignment in CLIP to Alleviate Single Tag Bias","arxiv_id":"2404.00384","date":"2024-03-30","proceeding":null,"authors":["Sanghyun Jo","Soohyun Ryu","Sungyub Kim","Eunho Yang","KyungSu Kim"],"abstract":"We identify a critical bias in contemporary CLIP-based models, which we denote as single tag bias. This bias manifests as a disproportionate focus on a singular tag (word) while neglecting other pertinent tags, stemming from CLIP's text embeddings that prioritize one specific tag in image-text relationships. When deconstructing text into individual tags, only one tag tends to have high relevancy with CLIP's image embedding, leading to biased tag relevancy. In this paper, we introduce a novel two-step fine-tuning approach, Text-Tag Self-Distillation (TTD), to address this challenge. TTD first extracts image-relevant tags from text based on their similarity to the nearest pixels then employs a self-distillation strategy to align combined masks with the text-derived mask. This approach ensures the unbiased image-text alignment of the CLIP-based models using only image-text pairs without necessitating additional supervision. Our technique demonstrates model-agnostic improvements in multi-tag classification and segmentation tasks, surpassing competing methods that rely on external resources. The code is available at https://github.com/shjo-april/TTD.","url_abs":"https://arxiv.org/abs/2404.00384v2","url_pdf":"https://arxiv.org/pdf/2404.00384v2.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":"ttd-text-tag-self-distillation-enhancing","repo_url":"https://github.com/shjo-april/TTD","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"multi-label-text-classification","task_name":"Multi-Label Text Classification"},{"task_slug":"open-vocabulary-semantic-segmentation","task_name":"Open Vocabulary Semantic Segmentation"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":"tag","task_name":"TAG"},{"task_slug":"unsupervised-semantic-segmentation-with","task_name":"Unsupervised Semantic Segmentation with Language-image Pre-training"}],"methods":[{"method_slug":"align","method_name":"ALIGN"},{"method_slug":"focus","method_name":"Focus"}],"datasets_introduced":[{"slug":"cc3m-tagmask","name":"CC3M-TagMask","full_name":""}],"methods_introduced":[],"results":[{"leaderboard":"/sota/multi-label-text-classification-on-cc3m","task":"Multi-Label Text Classification","dataset":"CC3M-TagMask","model":"TTD (w/ fine-tuning)","rank_in_archive_order":1,"of":6,"metrics":{"Accuracy":"88.6","F1":"82.8","Precision":"88.3","Recall":"78.0","mAP":"93.7"},"uses_additional_data":false},{"leaderboard":"/sota/multi-label-text-classification-on-cc3m","task":"Multi-Label Text Classification","dataset":"CC3M-TagMask","model":"TTD (w/o fine-tuning)","rank_in_archive_order":2,"of":6,"metrics":{"Accuracy":"91.0","F1":"78.5","Precision":"82.9","Recall":"74.5","mAP":"90.3"},"uses_additional_data":false},{"leaderboard":"/sota/open-vocabulary-semantic-segmentation-on-2","task":"Open Vocabulary Semantic Segmentation","dataset":"ADE20K-150","model":"TTD (TCL)","rank_in_archive_order":21,"of":23,"metrics":{"mIoU":"17.0"},"uses_additional_data":false},{"leaderboard":"/sota/open-vocabulary-semantic-segmentation-on-2","task":"Open Vocabulary Semantic Segmentation","dataset":"ADE20K-150","model":"TTD (MaskCLIP)","rank_in_archive_order":23,"of":23,"metrics":{"mIoU":"12.7"},"uses_additional_data":false},{"leaderboard":"/sota/open-vocabulary-semantic-segmentation-on-coco","task":"Open Vocabulary Semantic Segmentation","dataset":"COCO-Stuff-171","model":"TTD (TCL)","rank_in_archive_order":4,"of":7,"metrics":{"mIoU":"23.7"},"uses_additional_data":false},{"leaderboard":"/sota/open-vocabulary-semantic-segmentation-on-coco","task":"Open Vocabulary Semantic Segmentation","dataset":"COCO-Stuff-171","model":"TTD (MaskCLIP)","rank_in_archive_order":7,"of":7,"metrics":{"mIoU":"19.4"},"uses_additional_data":false},{"leaderboard":"/sota/open-vocabulary-semantic-segmentation-on","task":"Open Vocabulary Semantic Segmentation","dataset":"Cityscapes","model":"TTD (TCL)","rank_in_archive_order":3,"of":5,"metrics":{"mIoU":"32.0"},"uses_additional_data":false},{"leaderboard":"/sota/open-vocabulary-semantic-segmentation-on","task":"Open Vocabulary Semantic Segmentation","dataset":"Cityscapes","model":"TTD (MaskCLIP)","rank_in_archive_order":5,"of":5,"metrics":{"mIoU":"27.0"},"uses_additional_data":false},{"leaderboard":"/sota/open-vocabulary-semantic-segmentation-on-1","task":"Open Vocabulary Semantic Segmentation","dataset":"PASCAL Context-59","model":"TTD (TCL)","rank_in_archive_order":20,"of":24,"metrics":{"mIoU":"37.4"},"uses_additional_data":false},{"leaderboard":"/sota/open-vocabulary-semantic-segmentation-on-1","task":"Open Vocabulary Semantic Segmentation","dataset":"PASCAL Context-59","model":"TTD (MaskCLIP)","rank_in_archive_order":23,"of":24,"metrics":{"mIoU":"31.0"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-segmentation-on-cc3m-tagmask","task":"Semantic Segmentation","dataset":"CC3M-TagMask","model":"TTD (TCL)","rank_in_archive_order":1,"of":4,"metrics":{"mIoU":"65.5"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-segmentation-on-cc3m-tagmask","task":"Semantic Segmentation","dataset":"CC3M-TagMask","model":"TTD (MaskCLIP)","rank_in_archive_order":3,"of":4,"metrics":{"mIoU":"50.2"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-4","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"ADE20K","model":"TTD (TCL)","rank_in_archive_order":8,"of":13,"metrics":{"Mean IoU (val)":"17.0"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-4","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"ADE20K","model":"TTD (MaskCLIP)","rank_in_archive_order":10,"of":13,"metrics":{"Mean IoU (val)":"12.7"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-10","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"COCO-Object","model":"TTD (TCL)","rank_in_archive_order":4,"of":12,"metrics":{"mIoU":"37.4"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-10","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"COCO-Object","model":"TTD (MaskCLIP)","rank_in_archive_order":10,"of":12,"metrics":{"mIoU":"26.5"},"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":"TTD (TCL)","rank_in_archive_order":6,"of":12,"metrics":{"mIoU":"23.7"},"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":"TTD (MaskCLIP)","rank_in_archive_order":9,"of":12,"metrics":{"mIoU":"19.4"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-3","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"Cityscapes val","model":"TTD (MaskCLIP)","rank_in_archive_order":5,"of":12,"metrics":{"mIoU":"32.0"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-3","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"Cityscapes val","model":"TTD (TCL)","rank_in_archive_order":7,"of":12,"metrics":{"mIoU":"27.0"},"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":"TTD (TCL)","rank_in_archive_order":6,"of":12,"metrics":{"mIoU":"37.4"},"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":"TTD (MaskCLIP)","rank_in_archive_order":9,"of":12,"metrics":{"mIoU":"31.0"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-11","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"PASCAL VOC","model":"TTD (TCL)","rank_in_archive_order":6,"of":10,"metrics":{"mIoU":"61.1"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-semantic-segmentation-with-11","task":"Unsupervised Semantic Segmentation with Language-image Pre-training","dataset":"PASCAL VOC","model":"TTD (MaskCLIP)","rank_in_archive_order":9,"of":10,"metrics":{"mIoU":"43.1"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2404.00384","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2404.00384"}},"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/shjo-april/TTD","reach":null}],"summary":{"ran_draft_wrong":1},"by_repo_kind":{"official":{"samples":1,"ran":1,"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":1,"samples":[{"code_sha256_prefix":"6901388a150b1941","entry":"load","repo":"shjo-april/TTD","repo_kind":"official","path":"core/tcl/clip/clip.py","file_url":"https://github.com/shjo-april/TTD/blob/HEAD/core/tcl/clip/clip.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"6901388a150b1941"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}