{"paper":{"title":"InterPartAbility: Phrase-Region Grounding for Interpretable Text-to-Image Person Re-Identification","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"InterPartAbility adds text-guided part matching and constrained attention to produce grounded explanations for text-to-image person re-identification.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Aryan Shukla, Eric Granger, Maguelonne Heritier, Rajarshi Bhattacharya, Shakeeb Murtaza","submitted_at":"2026-04-29T19:18:36Z","abstract_excerpt":"Text-to-image person re-identification (TI-ReID) relies on natural-language text descriptions to retrieve top matching individuals from a gallery of reference images. While recent large vision-language models (VLMs) achieve strong retrieval performance, their decisions remain largely uninterpretable. Existing interpretability approaches in TI-ReID rely solely on slot-attention to highlight attended regions, but fail to reliably bind visual regions to semantically meaningful concepts, limiting interpretation to qualitative visualizations over a restricted vocabulary. This paper introduces Inter"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Empirical results on challenging benchmarks like CUHK-PEDES and ICFG-PEDES show that InterPartAbility achieves state-of-the-art (SOTA) interpretability performance under these metrics, while sustaining competitive retrieval accuracy.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the patch-phrase interaction module and constrained CLIP ViT self-attention reliably bind visual regions to semantically meaningful part phrases, which may not hold without further validation of the grounding quality.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"InterPartAbility performs explicit part-wise matching in text-to-image person re-identification via a patch-phrase interaction module to produce grounded explanations and achieves SOTA interpretability scores while maintaining competitive retrieval accuracy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"InterPartAbility adds text-guided part matching and constrained attention to produce grounded explanations for text-to-image person re-identification.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"fd025ce786d0f9c4ececfdb396ee0f8a9e3a8bceac8c61bc052a49d30751550c"},"source":{"id":"2604.27122","kind":"arxiv","version":2},"verdict":{"id":"41d31c4f-1c71-4e4f-97ea-ca8d472c4539","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T09:25:00.195583Z","strongest_claim":"Empirical results on challenging benchmarks like CUHK-PEDES and ICFG-PEDES show that InterPartAbility achieves state-of-the-art (SOTA) interpretability performance under these metrics, while sustaining competitive retrieval accuracy.","one_line_summary":"InterPartAbility performs explicit part-wise matching in text-to-image person re-identification via a patch-phrase interaction module to produce grounded explanations and achieves SOTA interpretability scores while maintaining competitive retrieval accuracy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the patch-phrase interaction module and constrained CLIP ViT self-attention reliably bind visual regions to semantically meaningful part phrases, which may not hold without further validation of the grounding quality.","pith_extraction_headline":"InterPartAbility adds text-guided part matching and constrained attention to produce grounded explanations for text-to-image person re-identification."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.27122/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T22:43:59.556510Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T19:33:34.214487Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"84072496d4463b0677dc00e59a503628e24a4f9f400a6f2c69d38bc4704c6695"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}