Pre-trained model is not available

[Xiaomin-Li]

Hi, I downloaded your pre-trained model from google drive. But it gives an error when loading the checkpoint.

----> 1 ckp = torch.load('./checkpoint_ssd300.pth.tar')

~/anaconda3/envs/PyTorch/lib/python3.8/site-packages/torch/serialization.py in load(f, map_location, pickle_module, **pickle_load_args)
591 return torch.jit.load(opened_file)
592 return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args)
--> 593 return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
594
595

~/anaconda3/envs/PyTorch/lib/python3.8/site-packages/torch/serialization.py in _legacy_load(f, map_location, pickle_module, **pickle_load_args)
770 unpickler = pickle_module.Unpickler(f, **pickle_load_args)
771 unpickler.persistent_load = persistent_load
--> 772 result = unpickler.load()
773
774 deserialized_storage_keys = pickle_module.load(f, **pickle_load_args)

ModuleNotFoundError: No module named 'model'

[Ryan-Holmes-363]

You need to extract the zip first (I think) but I'm having issues with an archive header when I try to do so.

[sgrvinod]

You need to extract the zip first (I think) but I'm having issues with an archive header when I try to do so.

Hi, you do not need to extract it first; you've to load it directly with PyTorch as written in the code. I'm sorry about the misleading name (".tar") of the file, it's a naming convention that I picked up somewhere that is understandably confusing.

I think the first commenter had that issue because the checkpoint contains the model saved directly in its full form, i. e. not only the state/parameters (state_dict()) of the model. Therefore, it requires that the file containing the model definition, model.py, is in the same working directory when you load the checkpoint.