1x-technologies/worldmodel

1X World Model Challenge数据集是为1X World Model Challenge设计的，包含训练和验证数据集，分为100个独立的分片。每个分片包含视频数据（8x8x8图像块，30hz，17帧时间窗口）、状态数据（np.float32格式）和元数据（metadata.json）。数据集的变化包括新增了100小时的训练和验证数据集，应用了面部模糊处理，并共享了一个新的原始视频数据集。此外，还提供了示例脚本来解码Cosmos Tokenized bins和加载帧数据。

The 1X World Model Challenge Dataset is designed for the 1X World Model Challenge, containing training and validation datasets divided into 100 independent shards. Each shard includes video data (8x8x8 image patches at 30hz with a 17-frame temporal window), state data (in np.float32 format), and metadata (metadata.json). Changes to the dataset include the addition of 100 hours of training and validation data, the application of face blurring, and the sharing of a new raw video dataset. Additionally, example scripts are provided to decode Cosmos Tokenized bins and load frame data.

原始信息汇总

数据集概述

数据集名称

1X World Model Challenge Dataset

许可证

Apache-2.0

数据集大小

10M < n < 100M

下载命令

shell huggingface-cli download 1x-technologies/worldmodel --repo-type dataset --local-dir data