18,187 images of 24,312 people, classified by annotators in 3 self-contact classes.
- TRAIN set: 16,329 images; 21,769 pairs of people classified to be in:
- no self-contact: 6,351 people
- uncertain self-contact: 8,258 people
- self-contact: 7,160 people
- TEST set: 1,858 images; 2,543 people classified
3,415 images of 3,969 people in self-contact with annotated self-contact signatures.
Each annotation represents a set of anntoated correspondences between 2 contact surfaces on the human body (self-contact signature) and their projection in the image (self-contact image support).
Signature correspondences are provideded on both GHUM and SMPLX templates, between:
- TRAIN set: 3,082 images; 3,568 people
- TEST set: 333 images; 401 people
- Vertex IDs
- Facet IDs
- Body region IDs (we split surfaces into 75 regions)