UltraVR benchmark evaluates VLMs on ultra-resolution image VQA

Arxiv CS.CV (computer vision)

The benchmark tests vision-language models on ultra-resolution images where critical evidence is tiny, subtle, or distributed. It aims to expose limitations in current models on high-resolution, evidence-grounded reasoning tasks.

vision-language-modelsbenchmarks

Arxiv CS.CV (computer vision)·Gexin HuangYanting YangMyeongkyun KangBeidi ZhaoJun ZhouChen ZhouGang WangZu-hua GaoXiaoxiao Li

6 days ago