nyu-visionx/pisa-experiments
收藏Hugging Face2025-03-18 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/nyu-visionx/pisa-experiments
下载链接
链接失效反馈官方服务：
资源简介：
PisaBench数据集是一个包含现实世界和模拟视频的数据集，用于研究和训练视频扩散模型中的物理后训练效果。数据集包括361个展示物体下落任务的现实世界视频，以及用于理解sim2real转移的60个模拟测试视频。此外，还包括使用Google的Kubric生成的模拟物理视频训练数据，以及用于物理监督微调（PSFT）和对象奖励优化（ORO）的模型 checkpoints。
The PisaBench dataset is a collection of real-world and simulated videos for studying and training the effect of physics post-training in video diffusion models. The dataset includes 361 real-world videos demonstrating object dropping tasks, as well as 60 simulated test videos for understanding sim2real transfer. Additionally, it contains simulated physics training data generated using Googles Kubric, and model checkpoints for Physics Supervised Fine-Tuning (PSFT) and Object Reward Optimization (ORO).
提供机构：
nyu-visionx
