{"id":28439208,"url":"https://github.com/vchitect/taca","last_synced_at":"2025-06-28T23:32:04.282Z","repository":{"id":298227617,"uuid":"991431698","full_name":"Vchitect/TACA","owner":"Vchitect","description":"TACA: Rethinking Cross-Modal Interaction in Multimodal Diffusion Transformers","archived":false,"fork":false,"pushed_at":"2025-06-10T02:58:30.000Z","size":105355,"stargazers_count":6,"open_issues_count":0,"forks_count":1,"subscribers_count":0,"default_branch":"main","last_synced_at":"2025-06-10T03:33:19.507Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/Vchitect.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-05-27T16:02:07.000Z","updated_at":"2025-06-10T03:28:44.000Z","dependencies_parsed_at":"2025-06-10T03:43:28.786Z","dependency_job_id":null,"html_url":"https://github.com/Vchitect/TACA","commit_stats":null,"previous_names":["vchitect/taca"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/Vchitect/TACA","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Vchitect%2FTACA","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Vchitect%2FTACA/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Vchitect%2FTACA/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Vchitect%2FTACA/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/Vchitect","download_url":"https://codeload.github.com/Vchitect/TACA/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Vchitect%2FTACA/sbom","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":262513619,"owners_count":23322661,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2025-06-06T02:07:53.938Z","updated_at":"2025-06-28T23:32:04.269Z","avatar_url":"https://github.com/Vchitect.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n\u003ch1\u003eTACA: Rethinking Cross-Modal Interaction in Multimodal Diffusion Transformers\u003c/h1\u003e\n\u003c/div\u003e\n\n\u003cdiv align=\"center\"\u003e\n    \u003cspan class=\"author-block\"\u003e\n      \u003ca href=\"https://scholar.google.com/citations?user=FkkaUgwAAAAJ\u0026hl=en\" target=\"_blank\"\u003eZhengyao Lv*\u003c/a\u003e\u003csup\u003e1\u003c/sup\u003e,\u003c/span\u003e\n    \u003c/span\u003e\n    \u003cspan class=\"author-block\"\u003e\n      \u003ca href=\"https://tianlinn.com/\" target=\"_blank\"\u003eTianlin Pan*\u003c/a\u003e\u003csup\u003e2,3\u003c/sup\u003e,\u003c/span\u003e\n    \u003c/span\u003e\n    \u003cspan class=\"author-block\"\u003e\n      \u003ca href=\"https://chenyangsi.github.io/\" target=\"_blank\"\u003eChenyang Si\u003c/a\u003e\u003csup\u003e2‡†\u003c/sup\u003e,\u003c/span\u003e\n    \u003c/span\u003e\n    \u003cspan class=\"author-block\"\u003e\n      \u003ca href=\"https://frozenburning.github.io/\" target=\"_blank\"\u003eZhaoxi Chen\u003c/a\u003e\u003csup\u003e4\u003c/sup\u003e,\u003c/span\u003e\n    \u003c/span\u003e\n    \u003cspan class=\"author-block\"\u003e\n      \u003ca href=\"https://homepage.hit.edu.cn/wangmengzuo\" target=\"_blank\"\u003eWangmeng Zuo\u003c/a\u003e\u003csup\u003e5\u003c/sup\u003e,\u003c/span\u003e\n    \u003c/span\u003e\n    \u003cspan class=\"author-block\"\u003e\n      \u003ca href=\"https://liuziwei7.github.io/\" target=\"_blank\"\u003eZiwei Liu\u003c/a\u003e\u003csup\u003e4†\u003c/sup\u003e,\u003c/span\u003e\n    \u003c/span\u003e\n    \u003cspan class=\"author-block\"\u003e\n      \u003ca href=\"https://i.cs.hku.hk/~kykwong/\" target=\"_blank\"\u003eKwan-Yee K. Wong\u003c/a\u003e\u003csup\u003e1†\u003c/sup\u003e\n    \u003c/span\u003e\n\u003c/div\u003e\n\n\u003cdiv align=\"center\"\u003e\n    \u003csup\u003e1\u003c/sup\u003eThe University of Hong Kong \u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; \n    \u003csup\u003e2\u003c/sup\u003eNanjing University \u003cbr\u003e \n    \u003csup\u003e3\u003c/sup\u003eUniversity of Chinese Academy of Sciences \u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; \n    \u003csup\u003e4\u003c/sup\u003eNanyang Technological University\u003cbr\u003e \n    \u003csup\u003e5\u003c/sup\u003eHarbin Institute of Technology\n\u003c/div\u003e\n\u003cdiv align=\"center\"\u003e(*Equal Contribution.\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u003csup\u003e‡\u003c/sup\u003eProject Leader.\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u003csup\u003e†\u003c/sup\u003eCorresponding Author.)\u003c/div\u003e\n\n\u003cp align=\"center\"\u003e\n    \u003ca href=\"https://arxiv.org/abs/2506.07986/\"\u003ePaper\u003c/a\u003e | \n    \u003ca href=\"https://vchitect.github.io/TACA/\"\u003eProject Page\u003c/a\u003e |\n    \u003ca href=\"https://huggingface.co/ldiex/TACA/tree/main\"\u003eLoRA Weights\u003c/a\u003e\n\u003c/p\u003e\n\n# About\nWe propose **TACA**, a parameter-efficient method that dynamically rebalances cross-modal attention in multimodal diffusion transformers to improve text-image alignment.\n\nhttps://github.com/user-attachments/assets/ae15a853-ee99-4eee-b0fd-8f5f53c308f9\n\n# Usage\nFor Stable Diffusion 3.5, simply run:\n``` sh\npython infer/infer_sd3.py\n```\n\nFor FLUX.1, run:\n``` sh\npython infer/infer_flux.py\n```\n\n# Benchmark\nComparison of alignment evaluation on T2I-CompBench for FLUX.1-Dev-based and SD3.5-Medium-based models.\n\n| Model | Attribute Binding | | | Object Relationship | | Complex $\\uparrow$ |\n|---|---|---|---|---|---|---|\n| | Color $\\uparrow$ | Shape $\\uparrow$ | Texture $\\uparrow$ | Spatial $\\uparrow$ | Non-Spatial $\\uparrow$ | |\n| FLUX.1-Dev | 0.7678 | 0.5064 | 0.6756 | 0.2066 | 0.3035 | 0.4359 |\n| FLUX.1-Dev + TACA ($r = 64$) | **0.7843** | **0.5362** | **0.6872** | **0.2405** | 0.3041 | **0.4494** |\n| FLUX.1-Dev + TACA ($r = 16$) | 0.7842 | 0.5347 | 0.6814 | 0.2321 | **0.3046** | 0.4479 |\n| SD3.5-Medium | 0.7890 | 0.5770 | 0.7328 | 0.2087 | 0.3104 | 0.4441 |\n| SD3.5-Medium + TACA ($r = 64$) | **0.8074** | **0.5938** | **0.7522** | **0.2678** | 0.3106 | 0.4470 |\n| SD3.5-Medium + TACA ($r = 16$) | 0.7984 | 0.5834 | 0.7467 | 0.2374 | **0.3111** | **0.4505** |\n\n# Showcases\n![](static/images/short_1.png)\n![](static/images/short_2.png)\n![](static/images/long_1.png)\n![](static/images/long_2.png)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fvchitect%2Ftaca","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fvchitect%2Ftaca","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fvchitect%2Ftaca/lists"}