跳到正文
arXiv cs.AI· Jiacheng Ding, Yanliang Qi, Cong Guo, Peihua Mai·· 2 小时前AI 评分18

See, Think, Use:诊断视觉语言模型感知与推理失败的方法(已撤回)

See, Think, Use: Diagnosing Perception and Reasoning Failures in VLMs

AI 导读

一篇 arXiv 论文提出 See, Think, Use 协议,将视觉语言模型(VLM)的每个答案追溯到感知、推理或对感知的使用三个阶段,指导开发者定位失败环节并分配测试时算力。该协议每题追加两个问题,无需访问模型权重,在 2 个模型家族和 6 个数据集上验证有效。该论文已被作者撤回,理由是稿件经大幅修改用于会议投稿,早期版本不再反映当前工作范围。

正文 · 原文

This paper has been withdrawn by Cong Guo

No PDF available, click to view other formats

Abstract:We present See, Think, Use, a protocol that traces each answer of a vision-language model to perception, reasoning or the use of perception. The diagnosis shows a developer which stage to improve and directs test-time compute to the stage that fails. Benchmarks score the final answer, and an answer that rests on the image earns the same credit as a guess or a memorized prior. The protocol asks the model two extra questions per item and needs no access to its weights. A localization probe asks for a box around every instance and measures seeing, and a text-replacement probe answers from a description and measures thinking. Use is the rate at which the answer follows the model's own perception across counterfactual image pairs. The rate needs no answer labels, because the question and its prior are the same on both images. On two model families and six datasets the protocol shows that the knob that helps follows the task family. Answers follow the model's perception except on images that contradict a common prior, and asking the model to localize first repairs this failure when its boxes are right. Allocating compute by the measured stages comes close to the best uniform policy at lower cost and beats a random allocation of the same compute wherever a knob raises the short stage. We will release the code, the prompts and the per-question outputs.
Comments: The authors request withdrawal of this preprint. The manuscript has been substantially revised for submission to a conference, and this early version no longer reflects the scope and content of the current work
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
Cite as: arXiv:2611.01467 [cs.CV]
  (or arXiv:2611.01467v2 [cs.CV] for this version)
  https://doi.org/10.48550/arXiv.2611.01467

arXiv-issued DOI via DataCite

Submission history

From: Cong Guo [view email]
[v1] Sun, 1 Nov 2026 02:56:06 UTC (2,943 KB)
[v2] Tue, 17 Nov 2026 06:30:11 UTC (1 KB) (withdrawn)

来源:arXiv cs.AI · arxiv.org