search

hongsukchoi / TCMR_RELEASE

Official Pytorch implementation of "Beyond Static Features for Temporally Consistent 3D Human Pose and Shape from a Video", CVPR 2021
MIT License
287 stars 39 forks source link

size mismatch #16

Open headre opened 2 years ago

headre commented 2 years ago

Thanks for your great work! When I tested the demo provided, some problem occurred while loading models:

Traceback (most recent call last):
  File "demo.py", line 376, in <module>
    main(args)
  File "demo.py", line 111, in main
    model.load_state_dict(ckpt, strict=False)
  File "/opt/data/private/TCMR/TCMR_RELEASE/tcmr-env/lib/python3.7/site-packages/torch/nn/modules/module.py", line 830, in load_state_dict
    self.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for TCMR:
    size mismatch for regressor.smpl.shapedirs: copying a param with shape torch.Size([6890, 3, 10]) from checkpoint, the shape in current model is torch.Size([6890, 3, 300]).

I wonder why and what to do. THX.

hongsukchoi commented 2 years ago

Hi, and sorry for the late reply...

I think the error is due to version matching of SMPL weights. Refer to this issue! https://githubmemory.com/repo/Jeff-sjtu/HybrIK/issues/24

fengzehui0422 commented 8 months ago

Hi, and sorry for the late reply...

I think the error is due to version matching of SMPL weights. Refer to this issue! https://githubmemory.com/repo/Jeff-sjtu/HybrIK/issues/24

sorry，the link is closed how can I solve this ？THX