{"id":24396796,"url":"https://github.com/ai4ce/llm4vpr","last_synced_at":"2025-04-11T16:41:48.532Z","repository":{"id":245432263,"uuid":"818075690","full_name":"ai4ce/LLM4VPR","owner":"ai4ce","description":"Can multimodal LLM help visual place recognition?","archived":false,"fork":false,"pushed_at":"2024-06-26T07:37:04.000Z","size":8300,"stargazers_count":37,"open_issues_count":0,"forks_count":1,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-03-25T12:51:18.411Z","etag":null,"topics":["llm","robotics","vision-and-language","vision-language-model","visual-place-recognition","vpr"],"latest_commit_sha":null,"homepage":"https://ai4ce.github.io/LLM4VPR/","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/ai4ce.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}},"created_at":"2024-06-21T03:48:53.000Z","updated_at":"2025-03-04T14:04:31.000Z","dependencies_parsed_at":"2024-06-22T04:35:51.747Z","dependency_job_id":"43c0e603-8382-48d2-ac94-08ffcb62f5ee","html_url":"https://github.com/ai4ce/LLM4VPR","commit_stats":null,"previous_names":["ai4ce/llm4vpr"],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ai4ce%2FLLM4VPR","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ai4ce%2FLLM4VPR/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ai4ce%2FLLM4VPR/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ai4ce%2FLLM4VPR/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/ai4ce","download_url":"https://codeload.github.com/ai4ce/LLM4VPR/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":248442236,"owners_count":21104138,"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":["llm","robotics","vision-and-language","vision-language-model","visual-place-recognition","vpr"],"created_at":"2025-01-19T21:58:29.089Z","updated_at":"2025-04-11T16:41:48.510Z","avatar_url":"https://github.com/ai4ce.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"## *Tell Me Where You Are*: Multimodal LLMs Meet Place Recognition\n[Zonglin Lyu](https://zonglinl.github.io/), [Juexiao Zhang](https://juexzz.github.io/), [Mingxuan Lu](https://scholar.google.com/citations?user=m4ChlREAAAAJ\u0026hl=en), [Yiming Li](https://yimingli-page.github.io/), [Chen Feng](https://ai4ce.github.io/)\n\n![image](./misc/images/Teaser.jpg)\n\n### Abstract\n\nLarge language models (LLMs) exhibit a variety of promising capabilities in robotics, \nincluding long-horizon planning and commonsense reasoning. \nHowever, their performance in place recognition is still underexplored. \nIn this work, we introduce multimodal LLMs (MLLMs) to visual place recognition (VPR), \nwhere a robot must localize itself using visual observations. \nOur key design is to use *vision-based retrieval* to propose several candidates and then leverage *language-based reasoning*\nto carefully inspect each candidate for a final decision. \nSpecifically, we leverage the robust visual features produced by off-the-shelf vision foundation models (VFMs) to obtain several candidate locations. \nWe then prompt an MLLM to describe the differences between the current observation and each candidate in a pairwise manner, \nand reason about the best candidate based on these descriptions.  Our method is termed **LLM-VPR**.\nResults on three datasets demonstrate that integrating the *general-purpose visual features* from VFMs with the *reasoning capabilities* of MLLMs \nalready provides an effective place recognition solution, *without any VPR-specific supervised training*. \nWe believe LLM-VPR can inspire new possibilities for applying and designing foundation models, i.e. VFMs, LLMs, and MLLMs, \nto enhance the localization and navigation of mobile robots.\n\n![image](./misc/images/LLM-VPR.jpg)\n\n**🔍  Please check out [project website](https://ai4ce.github.io/LLM4VPR/) for more details.**\n\n### Datasets\n\nPlease refer to [Anyloc](https://github.com/AnyLoc/AnyLoc) for dataset download. We included Baidu Mall, Pittsburgh30K, and Tokyo247.\n\n### Vision Foundation Model\n\nPlease refer to [Anyloc](https://github.com/AnyLoc/AnyLoc) for Vision Foundation Model. We employed DINO-v2-GeM in their setup. \n\nSave your Coarse retrieval results as the followng structures:\n\n```\n└──── \u003cworking directory\u003e/\n    ├──── \u003cdata_name\u003e/\n    |   ├──── Query.png\n    |   ├──── Top1_True/False.png\n    |   ├──── ...\n```\nIf the retrieval results is correct, the set is as True, otherwise False. This **does not** intend to tell MLLMs about the results. Instead, it is easier for you to compute whether MLLM imrpoves the performance or not. The True/False will be removed when they are fed to the MLLM.\n\n### Try Vision-Language Refiner\n\nWrite your own api keys and change the directory of the saved data in ```main.py``` and run:\n\n```\npython main.py\n```\n\nThis will generate .txt files with descriptions and reasonings. It will also provide the reranked Top-K by printing it to the terminal.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fai4ce%2Fllm4vpr","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fai4ce%2Fllm4vpr","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fai4ce%2Fllm4vpr/lists"}