{"paper":{"title":"Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"CAFT aligns local descriptions in long captions to image regions before forming global scene representations.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Byeonghyun Pak, Byeongju Woo, Sangwoo Mo, Stella X. Yu, Zilin Wang","submitted_at":"2026-02-03T01:31:55Z","abstract_excerpt":"Vision-language models such as CLIP often struggle to faithfully understand long, detail-rich captions, relying on dominant scene cues while overlooking fine-grained visual evidence. We propose a hierarchical vision-language learning principle for understanding scenes as part-to-whole compositions: before forming a whole-scene representation, a model should uncover what semantic parts appear where in the image. To this end, we propose CAFT (Cross-domain Alignment of Forests and Trees), a vision-language model that jointly learns local text-region alignment at intermediate representations and g"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"CAFT achieves state-of-the-art performance on six long-text retrieval benchmarks and exhibits strong scaling behavior. Experiments show that CAFT learns fine-grained representations that localize textual semantics in image regions without explicit region-level supervision.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that long captions naturally contain local descriptions that correspond to distinct scene parts, allowing the model to discover localized alignments without any region-level supervision or explicit part annotations.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CAFT achieves state-of-the-art results on long-text image retrieval benchmarks by jointly learning local text-region alignments and global image-text alignments through fine-to-coarse encoders.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"CAFT aligns local descriptions in long captions to image regions before forming global scene representations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"cdafd1a4fb73a5b2628f2b19b232934e702e23c6aa45d35cd110fdaa821467e7"},"source":{"id":"2602.02977","kind":"arxiv","version":2},"verdict":{"id":"ba226ff3-e405-4282-b468-ec9830987666","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T08:42:39.275338Z","strongest_claim":"CAFT achieves state-of-the-art performance on six long-text retrieval benchmarks and exhibits strong scaling behavior. Experiments show that CAFT learns fine-grained representations that localize textual semantics in image regions without explicit region-level supervision.","one_line_summary":"CAFT achieves state-of-the-art results on long-text image retrieval benchmarks by jointly learning local text-region alignments and global image-text alignments through fine-to-coarse encoders.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that long captions naturally contain local descriptions that correspond to distinct scene parts, allowing the model to discover localized alignments without any region-level supervision or explicit part annotations.","pith_extraction_headline":"CAFT aligns local descriptions in long captions to image regions before forming global scene representations."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"a349154a9bcf3d230e6f0fe39dc318a25a9a21e1bacd683f5d954d6b1f9e92dd"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}