Saliency-R1：通过显著性图对齐奖励强化可解释且忠实的视觉语言推理
Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Reward
提出一种无需额外计算开销的显著性图方法，可追踪视觉信息在推理链中的流动，并用显著性图与人工标注框的重叠度作为 GRPO 奖励，为 VLM 的视觉 grounding 与忠实性训练提供了可复用的奖励设计与归因工具。
原文依据 We use the overlap between the saliency maps and human-annotated bounding boxes as the reward function
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论视觉语言模型（VLM）在多种任务上取得了显著成功。然而，关于其可信性的担忧依然存在，尤其是它们倾向于更多依赖文本线索而非视觉证据，以及产生无依据或虚构回答的风险。为解决这些问题，我们提出 Saliency-R1，一个用于提升 VLM 推理可解释性与忠实性的框架。具体而言，我们引入一种新颖的显著性图技术，能够高效地突出对生成 token 有贡献的关键图像区域，且不带来额外计算开销。该技术可进一步扩展用于追踪视觉信息如何经由推理过程流向最终答案，从而揭示思维过程与视觉上下文之间的对齐关系。我们以显著性图与人工标注边界框之间的重叠度作为奖励函数，并应用组相对策略优化（GRPO）来对齐显著部分与关键区域，鼓励模型在推理时关注相关区域。实验表明，Saliency-R1 提升了推理的忠实性、可解释性以及整体任务性能。
英文原文摘要 1183 字符
Vision-language models (VLMs) have achieved remarkable success across diverse tasks. However, concerns about their trustworthiness persist, particularly regarding tendencies to lean more on textual cues than visual evidence and the risk of producing ungrounded or fabricated responses. To address these issues, we propose Saliency-R1, a framework for improving the interpretability and faithfulness of VLMs reasoning. Specifically, we introduce a novel saliency map technique that efficiently highlights critical image regions contributing to generated tokens without additional computational overhead. This can further be extended to trace how visual information flows through the reasoning process to the final answers, revealing the alignment between the thinking process and the visual context. We use the overlap between the saliency maps and human-annotated bounding boxes as the reward function, and apply Group Relative Policy Optimization (GRPO) to align the salient parts and critical regions, encouraging models to focus on relevant areas when conduct reasoning. Experiments show Saliency-R1 improves reasoning faithfulness, interpretability, and overall task performance.