Downloaded pre-trained weights cannot be loaded

[PeterKKan]

I downloaded weights that you provided from google drive(fold 1 best.pth) and try to use it in train process. here's the back trace.
loading model from best.pth
Traceback (most recent call last):
File "train_and_eval/segmentation_training_transf.py", line 221, in
train_and_evaluate(net, dataloaders, config, device)
File "train_and_eval/segmentation_training_transf.py", line 118, in train_and_evaluate
load_from_checkpoint(net, checkpoint, partial_restore=False)
File "/home/omnisky/kxb/TSVIT/DeepSatModels-main/utils/torch_utils.py", line 34, in load_from_checkpoint
net.load_state_dict(saved_net, strict=True)
File "/home/omnisky/.conda/envs/py38/lib/python3.8/site-packages/torch/nn/modules/module.py", line 2153, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for TSViT:
size mismatch for temporal_token: copying a param with shape torch.Size([1, 19, 128]) from checkpoint, the shape in current model is torch.Size([1, 20, 128]).

[michaeltrs]

Hi @PeterKKan,

From your error message I understand you are defining 20 output classes. This does not correspond to these pre-trained models which are trained for 19 classes. See the provided config files in https://github.com/michaeltrs/DeepSatModels/tree/main/configs/PASTIS24.