Back
Learning to Regress Bodies from Images using Differentiable Semantic Rendering
DSR uses semantic information in the form of clothing segmentation when learning to regress 3D human pose and shape from an image. The SMPL body should fit inside the clothing and match unclothed regions.