mks0601 / I2L-MeshNet_RELEASE

Official PyTorch implementation of "I2L-MeshNet: Image-to-Lixel Prediction Network for Accurate 3D Human Pose and Mesh Estimation from a Single RGB Image", ECCV 2020

MIT License

710 stars 130 forks source link

why not change the smpl_shape before self.smpl.layer['neutral'](smpl_pose, smpl_shape) #65

Closed zhLawliet closed 3 years ago

zhLawliet commented 3 years ago

smpl_mesh_coord, smpl_joint_coord = self.smpl.layer['neutral'](smpl_pose, smpl_shape)

change to mean shape if beta is too far from it

smpl_shape[(smpl_shape.abs() > 3).any(dim=1)] = 0.

if the beta is error,the smpl_joint_coord will be error. @mks0601

mks0601 commented 3 years ago

You are right. However, I found that there is a marginal performance difference.

zhLawliet commented 3 years ago

@mks0601 Will there be conflicts when using orig_joint_cam and fit_joint_cam together，I think it's better to just use fit_joint_cam , which fit for smpl. Have you done an experiment about the orig_joint_cam and fit_joint_cam.

mks0601 commented 3 years ago

fit_joint_cam provides smpl joint coordinates. However, they have some fitting errors.