Backbone mismatch between CASIA-B pre-trained model and model config.

[khoapham2801]

@tteepe I faced an issue when trying to run predict on CASIA-B, there were many unexpected layers not recognized when loading state_dict. The result of my investigation points out that there was a backend mismatch, more details below:

• In casia_b.yaml: backend_name: resgcn-n21-r8.
• However, the pre-trained model used backbone resgcn-n31-r8 as stated in hparams.yaml and the training log file of this model as well.
• There is no such n31 structure defined in the __init__.py of ResGCNv1, thus changing the backend_name solves nothing.

[ramprabhakar]

Hi @khoapham2801, I am also facing the same issue. How did you solve this problem? Thanks!