Soomro et al. (2012) Unconditional Video Generation Benchmark
收藏arXiv2025-09-30 收录
下载链接：
https://video-diffusion.github.io/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个基准数据集，包含了人们进行101种活动中的一段段短剪辑，这些数据最初是为了训练动作识别模型而收集的。模型评估的是每段16帧的短片段，这些片段被下采样至64x64的空间分辨率。该数据集规模包含13,320个视频，任务是对无条件的视频生成进行评估。
This is a benchmark dataset consisting of short video clips depicting people performing 101 distinct activities, which was initially collected for training action recognition models. Evaluations are conducted on 16-frame short clips, which are downsampled to a spatial resolution of 64×64. The dataset contains a total of 13,320 videos, and its designated task is to evaluate unconditional video generation.
提供机构：
TensorFlow Datasets
