Bad results with partial body

akashsengupta1997 / STRAPS-3DHumanShapePose

Code repository for the paper: Synthetic Training for Accurate 3D Human Pose and Shape Estimation in the Wild (BMVC 2020)

MIT License

165 stars 18 forks source link

Bad results with partial body #17

Closed hust-kevin closed 3 years ago

hust-kevin commented 3 years ago

hi, it seems get wrong result with partial body. There are some way or optimization to do to improve those cases

akashsengupta1997 commented 3 years ago

Hi, to predict 3D from partial body images you will need to train the network with partial body images. This can be done by simply adding in more significant occlusion/cropping/zooming augmentations such that only part of the the body is visible in the training silhouette+joint maps. I have gotten this to work OK in some other projects since this repo was released (see below).

akashsengupta1997 commented 3 years ago

See https://github.com/akashsengupta1997/HierarchicalProbabilistic3DHuman