Checkpoint file in readme doesn't work

[mjsampson]

OS: Ubuntu 16.04 Envrionment: docker from pytorch/pytorch:1.0-cuda10.0-cudnn7-runtime

Hello,

I am researching into view invariant 3d pose detection with the University of New Brunswick for it's applications in senior mobility and we are running your "in the wild" code! The link you have in your README no longer points to an active server and when I download the checkpoint file of the pretrained model from the source github account I get the following error

topiascz    | INFO: Trainable parameter count: 17004595
topiascz    | Loading checkpoint checkpoint/pretrained_h36m_cpn.bin
topiascz    | This model was trained for 80 epochs
topiascz    | Traceback (most recent call last):
topiascz    |   File "/project/subtree/tobiasc/run_wild.py", line 185, in <module>
topiascz    |     model_pos.load_state_dict(checkpoint['model_pos'])
topiascz    |   File "/opt/conda/lib/python3.6/site-packages/torch/nn/modules/module.py", line 769, in load_state_dict
topiascz    |     self.__class__.__name__, "\n\t".join(error_msgs)))
topiascz    | RuntimeError: Error(s) in loading state_dict for TemporalModel:
topiascz    |   size mismatch for expand_conv.weight: copying a param with shape torch.Size([1024, 34, 3]) from checkpoint, the shape in current model is torch.Size([1024, 51, 3]).
topiascz exited with code 1

Do you happen to have a link to the checkpoint file you used with your run_wild.py code. If not would simply editing the model.py file work?

Thankyou,

Matthew Sampson

[tobiascz]

I'm glad to hear that the code is used! correct weights file that should be the right one. You can find more Info here. I hope you don't have any further issues! I also updated the README with the modified links.

Cheers