arXiv Artificial Intelligence· Khai Le-Duc, Nguyen Kim Hai Bui, An Ngo, Nam D. Hoang, Yueqi Song, Jiarui Liu, Thai Quoc Hoang, Fang Wu, Chris Ngo, Terry Jingchen Zhang, Mona T. Diab, Ran Xu, Silvio Savarese, Xiao-Ming Wu·· 17 小时前AI 评分49
视觉语言模型在3D推理上为何失败？BedBench基准揭示精确计数短板
Where Do Vision Language Models Fail at 3D Reasoning?
AI 导读
研究团队推出BedBench基准，用13项程序化生成任务评测视觉语言模型的3D推理能力，发现按颜色精确计数时人类准确率93.3%，最佳模型仅7.2%。定向微调可将某模型准确率从3.0%提升至68.6%，但其对最常见颜色的回答仍与自身分色计数不一致。
来源：arXiv Artificial Intelligence · arxiv.org