InterPartAbility: Phrase-Region Grounding for Interpretable Text-to-Image Person Re-Identification cs.CV · 2026-04-29