English(EN) Small Vision-Language Models Know When They Are Wrong But Cannot Say So: A Two-Model Study of Stated versus Internal Confidence Under Realistic Image Degradation

小型视觉语言模型在图像退化下表现出置信度差距

作者 PulseAugur 编辑部 · [1 个来源] · 2026-07-27 04:00

一项近期研究考察了Qwen2-VL-2B-Instruct和SmolVLM-Instruct这两个小型视觉语言模型在真实图像退化下的置信度信号。研究发现，在自然语言中声明的置信度与内部标记概率之间存在显著差异。虽然内部概率被证明是更可靠的错误指标，但在面对严重曝光不足时，两个模型都难以准确评估其置信度，导致准确性大幅下降，而置信度信号变化甚微。 AI

影响凸显了当前小型视觉语言模型在可靠不确定性估计方面的局限性，这对于安全部署至关重要。

排序理由学术论文，详细介绍了模型能力的研究结果。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · M M Asif Ferdous · 2026-07-27 04:00

Small Vision-Language Models Know When They Are Wrong But Cannot Say So: A Two-Model Study of Stated versus Internal Confidence Under Realistic Image Degradation

arXiv:2607.22034v1 Announce Type: cross Abstract: Vision-language models (VLMs) are increasingly deployed on consumer hardware where input images are degraded by compression, camera shake, and poor lighting. In such settings, a reliable uncertainty signal matters more than raw ac…

报道来源 [1]

Small Vision-Language Models Know When They Are Wrong But Cannot Say So: A Two-Model Study of Stated versus Internal Confidence Under Realistic Image Degradation

相关实体

相关话题