Skip to content

About Training of Second Stage #2

@delCayr

Description

@delCayr

Hello, I am a master's student. I have successfully reproduced your work, and it is excellent. During the second training phase, I encountered the following error, corresponding to the code:

File "/home/cjm/RS3D/HJ/SAS/SAS-main/run/distill_EMA.py",
line 363, in distill history_pred = torch.load(os.path.join(path, scene_name + '.pth')).cuda(non_blocking=True).float()
File "/opt/conda/envs/SAS/lib/python3.8/site-packages/torch/serialization.py",
line 1028, in load return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
File "/opt/conda/envs/SAS/lib/python3.8/site-packages/torch/serialization.py",
line 1246, in _legacy_load magic_number = pickle_module.load(f, **pickle_load_args)
EOFError: Ran out of input

Furthermore, I discovered that the EMA update failed to update the new_pred file due to dimension mismatch in pred_history/scene0291_02.pth:
File "/home/cjm/RS3D/HJ/SAS/SAS-main/run/distill_EMA.py", line 236, in main_worker
loss_train = distill(train_loader, model, optimizer, epoch)
File "/home/cjm/RS3D/HJ/SAS/SAS-main/run/distill_EMA.py", line 383, in distill
new_pred = 0.9999 * history_pred + 0.0001 * current_pred
RuntimeError: The size of tensor a (87558) must match the size of tensor b (231093) at non-singleton dimension 0

Could you please help me? Thank you very much!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions