Entropy-Gradient Grounding: Training-Free Evidence Retrieval in Vision-Language Models
This addresses the challenge of improving evidence retrieval for vision-language models in tasks like document analysis and compositional queries, offering an incremental but practical enhancement without requiring retraining.
The paper tackled the problem of vision-language models struggling with queries requiring fine visual details or multi-region evidence by proposing a training-free grounding method that uses entropy gradients to retrieve relevant visual regions, achieving consistent improvements across seven benchmarks and four VLM architectures, with notable gains in detail-critical and high-resolution settings.
Despite rapid progress, pretrained vision-language models still struggle when answers depend on tiny visual details or on combining clues spread across multiple regions, as in documents and compositional queries. We address this by framing grounding as test-time evidence retrieval: given a query, the model should actively identify where to look next to resolve ambiguity. To this end, we propose a training-free, model-intrinsic grounding method that uses uncertainty as supervision. Specifically, we compute the entropy of the model's next-token distribution and backpropagate it to the visual token embeddings to obtain an entropy-gradient relevance map, without auxiliary detectors or attention-map heuristics. We then extract and rank multiple coherent regions to support multi-evidence queries, and introduce an iterative zoom-and-reground procedure with a spatial-entropy stopping rule to avoid over-refinement. Experiments on seven benchmarks across four VLM architectures demonstrate consistent improvements over existing methods, with the largest gains on detail-critical and high-resolution settings, while also producing more interpretable evidence localizations.