{"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/janus-decoupling-visual-encoding-for-unified","title":"Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation","arxiv_id":"2410.13848","date":"2024-10-17","proceeding":"CVPR 2025 1","authors":["Chengyue Wu","Xiaokang Chen","Zhiyu Wu","Yiyang Ma","Xingchao Liu","Zizheng Pan","Wen Liu","Zhenda Xie","Xingkai Yu","Chong Ruan","Ping Luo"],"abstract":"In this paper, we introduce Janus, an autoregressive framework that unifies multimodal understanding and generation. Prior research often relies on a single visual encoder for both tasks, such as Chameleon. However, due to the differing levels of information granularity required by multimodal understanding and generation, this approach can lead to suboptimal performance, particularly in multimodal understanding. To address this issue, we decouple visual encoding into separate pathways, while still leveraging a single, unified transformer architecture for processing. The decoupling not only alleviates the conflict between the visual encoder's roles in understanding and generation, but also enhances the framework's flexibility. For instance, both the multimodal understanding and generation components can independently select their most suitable encoding methods. Experiments show that Janus surpasses previous unified model and matches or exceeds the performance of task-specific models. The simplicity, high flexibility, and effectiveness of Janus make it a strong candidate for next-generation unified multimodal models.","url_abs":"https://arxiv.org/abs/2410.13848v1","url_pdf":"https://arxiv.org/pdf/2410.13848v1.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":"janus-decoupling-visual-encoding-for-unified","repo_url":"https://github.com/deepseek-ai/janus","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"visual-question-answering-1","task_name":"Visual Question Answering"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/on-wise","task":"","dataset":"WISE","model":"Janus","rank_in_archive_order":11,"of":11,"metrics":{"Biology":"0.28","Chemistry":"0.14","Cultural":"0.16","Overall":"0.23","Physics":"0.30","Space":"0.35","Time":"0.26"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-on-mm-vet","task":"Visual Question Answering","dataset":"MM-Vet","model":"Janus","rank_in_archive_order":168,"of":231,"metrics":{"GPT-4 score":"34.3"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2410.13848","atlas_url":"https://app.syntology.ai/?focus=2410.13848","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}