3DCP
Three Contact Poses

 

 

 

3DCP Scan: A dataset of body scans of five subjects (2 female, 3 male) in various self-contact poses with registrations in SMPL-X topology. In total, 3DCP Scan contains 586 3D body scans or 143/129/150/33/131 for subjects 00/01/02/03/04.

3DCP Mocap: We search for near-contact poses in the AMASS motion capture dataset. These meshes may contain slight self-interpenetration. To resolve this, we use our self-contact optimization method which maintains contact while resolving intersections. See GitHub for more information about the self-contact optimization. 3DCP Mocap pose parameters are included in the MTP dataset.

 

---

MTP
Mimic The Pose

 

The MTP Dataset has 3731 images from 148 different subjects, mimicking poses with self-contact sampled from 3DCP Scan, 3DCP Mocap and AGORA. With our SMPLify-XMC optimization method, we fit the SMPL-X to the mimicked images and obtain high quality reference poses.

The MTP dataset includes:
- mimicked images in original resolution
- subject metadata (gender, height, weight)
- the presented pose and contact parameters
- SMPLify-XMC fits (SMPL-X parameters)
- SMPLify-XMC fits converted to SMPL topology
- OpenPose keypoints

 

---

DSC
Discrete Self-Contact

 

 

DSC collect discrete self-contact annotations for images in DeepFasion, LSP and LSP Extended. To do this, we divide the body into 24 regions and annotate their pairwise contact. We collect 30K annotations in total (18K/2K/10K for DeepFasion/LSP/LSP Extended). Please download the images from original dataset websites.