World Model Tokenized Video Data For Compression Challenge

Approximately 100 hours of video data for the 1X World Model Compression Challenge, with faces blurred for privacy. The dataset includes a v2.0 train/validation split and a separate test set, provided by 1x-technologies and last updated in April 2025.

Use Cases

• Train video compression models on the v2.0 train/validation split of tokenized video data.
• Benchmark model performance on the separate v2.0 test dataset for the Compression Challenge.
• Apply privacy-preserving techniques to video data where faces are already blurred.

Strengths

• Contains approximately 100 hours of video data for model training and testing.
• Includes a dedicated v2.0 test dataset specifically for the Compression Challenge.
• Privacy is addressed through face blurring in the video content.
• Dataset was updated to v2.0 in April 2025, indicating recent maintenance.

Limitations

• The specific number of video clips, rows, columns, and file size are unknown.
• The exact tokenization method and data format are not described in the provided input.
• The dataset's license terms are not specified in the provided input.

Provenance

Source

1x-technologies

Collection Method

Created for the 1X World Model Compression Challenge; likely derived from raw video data.

Freshness

Last updated on 2025-04-20.