ClothTransformer: 2,056 Penetration-Free Cloth Simulation Trajectories

2,056 penetration-free cloth simulation trajectories, each 240 frames long, for a total of 493,440 frames. The dataset was created by YuCrazing1 as the official dataset for the ClothTransformer paper and was last updated on 2026-07-14. It contains cloth mesh interactions across three scenarios: free-fall collision, garments on animated human bodies, and robotic cloth manipulation.

Use Cases

• Training machine learning models for cloth dynamics based on the 493,440 frames of simulation data.
• Benchmarking cloth simulation algorithms based on the penetration-free trajectories across three defined scenarios.
• Developing robotic cloth manipulation policies based on the robotic manipulation scenario data.
• Creating realistic digital garments for animation based on the garments-on-human-bodies scenario.

Strengths

• Contains 2,056 distinct simulation trajectories, providing a substantial number of examples.
• Each trajectory is 240 frames long, resulting in a total of 493,440 frames for analysis.
• Data is explicitly penetration-free, a key quality metric for physical plausibility in simulation.
• Covers three distinct interaction scenarios, offering variety in application contexts.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for certain tasks.
• The dataset size is approximately 33 GB, which may require significant storage and bandwidth.

Provenance

Source

huggingface, uploaded by author YuCrazing1.

Collection Method

Generated as part of the ClothTransformer research project for scalable cloth simulation.

Freshness

Last updated 2026-07-14 05:53:42; freshness should be verified.