General-purpose Visual Understanding Evaluation (G-VUE)
收藏arXiv2022-11-28 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2211.15402v1
下载链接
链接失效反馈官方服务：
资源简介：
G-VUE是一个全面评估通用视觉理解能力的基准，包含11个精心挑选的任务，覆盖感知、基础、推理和行动四个功能域。这些任务从3D重建到视觉推理和操作，全面测试模型的视觉认知能力。G-VUE不仅提供了一个评估框架，还允许对任意视觉表示在所有11个任务上进行评估，旨在推动研究向构建通用视觉系统发展。
G-VUE is a benchmark for comprehensively evaluating general visual understanding capabilities. It includes 11 carefully selected tasks covering four functional domains: perception, foundational vision, reasoning, and action. These tasks range from 3D reconstruction to visual reasoning and manipulation, comprehensively testing the visual cognitive abilities of models. G-VUE not only provides an evaluation framework but also enables the evaluation of arbitrary visual representations across all 11 tasks, aiming to advance research toward the development of general-purpose visual systems.
提供机构：
北京通用人工智能研究院
创建时间：
2022-11-28
