Hi, thank you for your work.
Currently, I am training GauHuman on your dataset. I tried to reuse the Dataloader for ZJU, however the training result is always black. It seems that your dataset is using a different format with the ZJU Monocap dataset.
I only modify the data of camera translation by multiplying with camera_scale.pkl. The way I convert original data of MVHuman to ZJU format is:
new_params: taken from smpl/smplx.
new_vertices: taken from smpl_param
cams: taken from camera_intrinsics, camera_extrinsics
kpts2d: taken from openpose
Questions/notices:
- The SMPL parameters has shape of (1, 87), while it is (1, 72) in ZJU.
- Is the camera convention in MVHuman the same as in ZJU?
- How can I convert the MVHuman data to the exact format of ZJU?
Thank you!
Hi, thank you for your work.
Currently, I am training GauHuman on your dataset. I tried to reuse the Dataloader for ZJU, however the training result is always black. It seems that your dataset is using a different format with the ZJU Monocap dataset.
I only modify the data of camera translation by multiplying with
camera_scale.pkl. The way I convert original data of MVHuman to ZJU format is:new_params: taken fromsmpl/smplx.new_vertices: taken fromsmpl_paramcams: taken fromcamera_intrinsics,camera_extrinsicskpts2d: taken fromopenposeQuestions/notices:
Thank you!