CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation
[Widzenie komputerowe] Autorzy: Tingyu Song, Mingxin Li, Yanzhao Zhang i 5 więcej
MLLM-based embedding models remain limited in compositional retrieval, often failing to distinguish scenes containing the same concepts but different attribute-object bindings. Yet the same backbone can resolve such distinctions when used as a cross-attentive reranker, motivating us to distill its compositional judgments into the embedding model. We propose CORE, which synthesizes candidate lists spanning five compositional matching levels and introduces a Rank-KL objective that trains the embedding model to reproduce the reranker's fine-grained ranking. We further introduce a graded evaluation protocol and compare contrastive learning, pairwise CoSENT, and listwise Rank-KL under the same data and tuning budget. Our comparison shows that both CoSENT and Rank-KL use the multi-level supervision more effectively than contrastive learning, with Rank-KL achieving the strongest overall perfor…
- Kategoria arxiv
- cs.CV
- Kategoria
- Widzenie komputerowe