3DBodyTex.Pose offers high quality and rich data containing 405 different real subjects in various clothing and poses, and around 325k image samples with ground-truth 2D and 3D pose annotations. These images are generated from 800 viewpoints among which 480 challenging extreme viewpoints. The data were created starting from high resolution textured 3D body scans and by incorporating various realistic backgrounds. Retraining a state-of-the-art 3D pose estimation approach using 3DBodyTex.Pose as data augmentation, showed promising improvement in the overall performance. See the following table.
The 3DBodyTex.Pose dataset will be available soon for use by external parties.
- Towards Generalization of 3D Human Pose Estimation In The Wild, Baptista R., Saint A., Al Ismaeil K., Aouada D., (2020). 3D Human Understanding Workshop, ICPR 2020.