search

HongwenZhang / PyMAF-X

[TPAMI 2023] PyMAF-X: Towards Well-aligned Full-body Model Regression from Monocular Images
https://www.liuyebin.com/pymaf-x
Other
196 stars 27 forks source link

Regressing Hand Parameters only #7

Closed enesduran closed 1 year ago

enesduran commented 1 year ago

First, thank you for the great work and making it open source!

I want to regress solely hand shape, pose from video (translation and root orientation included). What should my running configuration be? I inspected the code but could not found.

Regards.

HongwenZhang commented 1 year ago

Hi, thanks for your interest in our work!

The results of PyMAF-X include the hand shape and pose, but it may need to modify several lines of the code.

For hand-only evaluation, the code can run as follows:

python -m apps.demo_smplx --image_folder examples/coco_images --detection_threshold 0.3 --pretrained_model output/PyMAF-X_model_checkpoint_v1.1.pt --misc TRAIN.BHF_MODE 'hand_only' MODEL.MESH_MODEL mano

PS: modify strict=True to strict=False as: model.load_state_dict(checkpoint['model'], strict= False) at https://github.com/HongwenZhang/PyMAF-X/blob/smplx/apps/demo_smplx.py#L199

Then the network should have the output dictionary as: https://github.com/HongwenZhang/PyMAF-X/blob/smplx/models/pymaf_net.py#L683

enesduran commented 1 year ago

Thank you for your help! Regards