Unverified50% confidenceFactExact time
In scientific chart question answering, VLMs must complete multi-level subtasks such as OCR, legend recognition, axis reading, numerical comparison, and arithmetic reasoning, and a deviation in any step can cause the final answer to mismatch the ground-truth answer.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedVLM训练需要精确对齐的图文配对数据集，若图文对应关系出现偏差，模型会学到错误的跨模态映射76% similarUnverifiedPhyMetric's evaluation is performed by a VLM as the judge, answering on question-aware sampled frames, after which accuracy is aggregated at four levels: scene, dimension, physical category, and model.74% similarUnverified多模态大模型（VLM）不是OCR，它在视觉向量上进行注意力计算并结合语言先验推断内容，对无规律哈希值会产生幻觉而非报识别失败72% similarUnverifiedVision-language models (VLMs) suffer from prompt bias: when the prompt explicitly specifies a target category, the model tends to 'find' objects matching the description in the image even when they do not actually exist.71% similarUnverifiedVGR needs to be manually designed or induced from data, and its quality determines the upper limit of the VLM judgment; if a certain aesthetic dimension is missed, the model will not optimize in that direction.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102166API
curl https://kongchang.com/api/v1/knowledge/claims/1102166MCP
get_claim(id=1102166)