How to train Joint regressor for a different dataset

hongsukchoi / Pose2Mesh_RELEASE

Official Pytorch implementation of "Pose2Mesh: Graph Convolutional Network for 3D Human Pose and Mesh Recovery from a 2D Human Pose", ECCV 2020

MIT License

661 stars 69 forks source link

How to train Joint regressor for a different dataset #72

Open zhipenggong opened 10 months ago

zhipenggong commented 10 months ago

Hi @hongsukchoi

I would like to train the model with some other datasets, could you please tell me how to train Joint regressor for a different dataset,

Thank you very much and looking forward to your reply

hongsukchoi commented 10 months ago

Hi, do you mean you want to train a PoseNet? What is the joint format of that dataset?

zhipenggong commented 10 months ago

Yes, it's a mocap dataset captured by us, but with different joints positions from human3.6m and coco. We would like to train a Joint regressor to convert SMPL model to the joints for the comparison. Could you please help give us some guidance? Thanks a lot.

hongsukchoi commented 9 months ago

Hi,

Sorry for the late reply. I remember we manually picking the location from the template posed mesh.