{"paper":{"title":"Ferret: Refer and Ground Anything Anywhere at Any Granularity","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Ferret unifies referring and grounding in multimodal LLMs via a hybrid region representation of coordinates and continuous features.","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Bowen Zhang, Haotian Zhang, Haoxuan You, Liangliang Cao, Shih-Fu Chang, Xianzhi Du, Yinfei Yang, Zhe Gan, Zirui Wang","submitted_at":"2023-10-11T17:55:15Z","abstract_excerpt":"We introduce Ferret, a new Multimodal Large Language Model (MLLM) capable of understanding spatial referring of any shape or granularity within an image and accurately grounding open-vocabulary descriptions. To unify referring and grounding in the LLM paradigm, Ferret employs a novel and powerful hybrid region representation that integrates discrete coordinates and continuous features jointly to represent a region in the image. To extract the continuous features of versatile regions, we propose a spatial-aware visual sampler, adept at handling varying sparsity across different shapes. Conseque"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The resulting model not only achieves superior performance in classical referring and grounding tasks, but also greatly outperforms existing MLLMs in region-based and localization-demanded multimodal chatting.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the spatial-aware visual sampler can reliably extract continuous features from regions of arbitrary shape and sparsity without introducing systematic bias or information loss that would affect downstream grounding accuracy.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Ferret introduces a hybrid region representation and the GRIT dataset to let MLLMs refer to and ground arbitrary image regions, outperforming prior models on referring, grounding, and localization-aware chatting while reducing object hallucination.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Ferret unifies referring and grounding in multimodal LLMs via a hybrid region representation of coordinates and continuous features.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"73a4e79f223b2e942c9dc25ab4e1a885e1134543daecb2348e8939682c1a03a5"},"source":{"id":"2310.07704","kind":"arxiv","version":1},"verdict":{"id":"657af406-82d9-4462-a844-036903d1320e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T11:24:46.089802Z","strongest_claim":"The resulting model not only achieves superior performance in classical referring and grounding tasks, but also greatly outperforms existing MLLMs in region-based and localization-demanded multimodal chatting.","one_line_summary":"Ferret introduces a hybrid region representation and the GRIT dataset to let MLLMs refer to and ground arbitrary image regions, outperforming prior models on referring, grounding, and localization-aware chatting while reducing object hallucination.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the spatial-aware visual sampler can reliably extract continuous features from regions of arbitrary shape and sparsity without introducing systematic bias or information loss that would affect downstream grounding accuracy.","pith_extraction_headline":"Ferret unifies referring and grounding in multimodal LLMs via a hybrid region representation of coordinates and continuous features."},"references":{"count":20,"sample":[{"doi":"","year":null,"title":"The length of the output list needs to be exactly equal to the input list","work_id":"5df37ad1-a803-4193-87c3-dab4bc212ed9","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Do not explain the reasons","work_id":"eff2168b-1779-4d8a-a995-c2290e9a4073","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Do not mention the input entities, at least the output name and input name needs to be different","work_id":"a08ad987-d7c3-4b1f-8e9d-7bc6fe9b579a","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Do not mention something abstract, like ¨alien¨","work_id":"c210e3f5-c95a-4540-a7cc-786189f56aac","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"When dealing with quantities, focus solely on increasing the numbers during revision","work_id":"b1ce2dc5-9cf0-4e5d-83d1-16bcf11a63cc","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":20,"snapshot_sha256":"a7ededc6035b60c8027ce87beee2e525dbb4657e9dec9803222aae38e0150bb4","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"135267e8810624bb106e0dc58d9ed73e308a9c2616ed8a4aa8c0277ed3d57860"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}