Section: New Software and Platforms

BodyNet

BodyNet: Volumetric Inference of 3D Human Body Shapes

Keywords: Computer vision - Machine learning

Functional Description: BodyNet has the code to train multi-task neural networks for 2D/3D pose estimation, 2D body part segmentation, and 3D volumetric shape estimation of human bodies given single RGB images as input. The release also contains pre-trained models.

• Participants: Gül Varol Simsekli, Ivan Laptev and Cordelia Schmid

• Contact: Gül Varol Simsekli

• Publication: BodyNet: Volumetric Inference of 3D Human Body Shapes

• URL: https://www.di.ens.fr/willow/research/bodynet/