GLIMPSE: Gradient-Layer Importance Mapping for Prompted Visual Saliency Explanation for Generative LVLMs

Abstract: Recent progress in large vision-LLMs (LVLMs) has advanced the state of the art in visual question answering (VQA). However, interpreting where LVLMs direct their visual attention while generating free-form responses remains a significant challenge, yet is essential for understanding model behavior. We introduce GLIMPSE (Gradient-Layer Importance Mapping for Prompted Visual Saliency Explanation), a lightweight, model-agnostic framework that jointly attributes LVLM outputs to the most relevant visual evidence and textual signals supporting open-ended VQA. GLIMPSE fuses gradient-weighted attention, adaptive layer propagation, and relevance-weighted token aggregation to produce holistic response-level heat maps for interpreting cross-modal reasoning, outperforming prior interpretability methods and pushing the state-of-the-art in human-alignment. We demonstrate an analytic explainable AI (XAI) approach using GLIMPSE to uncover fine-grained insights into LVLM cross-modal attribution, trace reasoning dynamics, analyze systematic human-attention misalignment, diagnose hallucination, expose bias, and ensure transparency.