VLM-Reasoning/VCR-Bench
收藏Hugging Face2025-05-11 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/VLM-Reasoning/VCR-Bench
下载链接
链接失效反馈官方服务：
资源简介：
VCR-Bench是一个用于评估大型视觉语言模型（LVLMs）视频链式思维推理能力的综合评估框架。该数据集通过7个不同的任务维度，全面覆盖多种视频类型和时长，并提供了详细的逐步推理注释。数据集由14个现有视频基准测试的数据组成，经过严格的手动注释和质控，包含了859个视频和1034个高质量的问答对。
VCR-Bench is a comprehensive evaluation framework designed to assess the video chain-of-thought reasoning capabilities of Large Vision-Language Models (LVLMs). The dataset covers a wide range of video types and durations across 7 distinct task dimensions, and provides detailed step-by-step reasoning annotations. It is compiled from 14 existing video benchmarks and has undergone rigorous manual annotation and quality control, resulting in a collection of 859 videos and 1,034 high-quality question-answer pairs.
提供机构：
VLM-Reasoning
