Cloth-splatters2026MIT
DexGarmentLab Lift Correspondence 2026-08-22
A point-cloud dataset of 1,800 successful garment lift/manipulation episodes with dense surface correspondence annotations, spanning tops, dresses, and trousers. Each episode contains up to 16 observations with 2,048 XYZ points and persistent surface-template IDs for cloth manipulation research.
Downloads0
Episodes1800
Why This Matters for Physical AI
This dense point-cloud correspondence dataset enables training of cloth state estimation and manipulation models by providing ground-truth surface tracking across deformation phases.
Technical Profile
- Modalities
- point_cloud
- Environment
- lab
- Task Types
- manipulationcloth-manipulationgrasping
- Episodes
- 1800
- Data Format
- NumPy
- Annotation Types
- semantic_keypoint_correspondencessurface_template_idsvisibility_distancesgrasp_phase_labels
- License
- MIT
Access
Need custom point_cloud data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack