search

toandaominh1997 / EfficientDet.Pytorch

Implementation EfficientDet: Scalable and Efficient Object Detection in PyTorch
MIT License
1.44k stars 306 forks source link

Update the url for the efficientnet backbone #110

Closed cristyioan2000 closed 4 years ago

cristyioan2000 commented 4 years ago

The owner of the efficientnet weights won't keep the weights on the google cloud because it's to expensive, you should upload or change theurl_map from utils.py with new urls to the weights.

Follow this issue for more info: https://github.com/lukemelas/EfficientNet-PyTorch/issues/138

Cyril9227 commented 4 years ago

I think in the model/utils.py file you should add the code below, also check my PR to see the other bits of code to change so everything is consistent with the new weights


url_map = {
    'efficientnet-b0': 'https://publicmodels.blob.core.windows.net/container/aa/efficientnet-b0-355c32eb.pth',
    'efficientnet-b1': 'https://publicmodels.blob.core.windows.net/container/aa/efficientnet-b1-f1951068.pth',
    'efficientnet-b2': 'https://publicmodels.blob.core.windows.net/container/aa/efficientnet-b2-8bb594d6.pth',
    'efficientnet-b3': 'https://publicmodels.blob.core.windows.net/container/aa/efficientnet-b3-5fb5a3c3.pth',
    'efficientnet-b4': 'https://publicmodels.blob.core.windows.net/container/aa/efficientnet-b4-6ed6700e.pth',
    'efficientnet-b5': 'https://publicmodels.blob.core.windows.net/container/aa/efficientnet-b5-b6417697.pth',
    'efficientnet-b6': 'https://publicmodels.blob.core.windows.net/container/aa/efficientnet-b6-c76e70fd.pth',
    'efficientnet-b7': 'https://publicmodels.blob.core.windows.net/container/aa/efficientnet-b7-dcc49843.pth',
}

url_map_advprop = {
    'efficientnet-b0': 'https://publicmodels.blob.core.windows.net/container/advprop/efficientnet-b0-b64d5a18.pth', 
    'efficientnet-b1': 'https://publicmodels.blob.core.windows.net/container/advprop/efficientnet-b1-0f3ce85a.pth',
    'efficientnet-b2': 'https://publicmodels.blob.core.windows.net/container/advprop/efficientnet-b2-6e9d97e5.pth',
    'efficientnet-b3': 'https://publicmodels.blob.core.windows.net/container/advprop/efficientnet-b3-cdd7c0f4.pth',
    'efficientnet-b4': 'https://publicmodels.blob.core.windows.net/container/advprop/efficientnet-b4-44fb3a87.pth',
    'efficientnet-b5': 'https://publicmodels.blob.core.windows.net/container/advprop/efficientnet-b5-86493f6b.pth',
    'efficientnet-b6': 'https://publicmodels.blob.core.windows.net/container/advprop/efficientnet-b6-ac80338e.pth',
    'efficientnet-b7': 'https://publicmodels.blob.core.windows.net/container/advprop/efficientnet-b7-4652b6dd.pth',
    'efficientnet-b8': 'https://publicmodels.blob.core.windows.net/container/advprop/efficientnet-b8-22a8fe65.pth',
}

def load_pretrained_weights(model, model_name, load_fc=True, advprop=False):
    """ Loads pretrained weights, and downloads if loading for the first time. """
    # AutoAugment or Advprop (different preprocessing)
    url_map_ = url_map_advprop if advprop else url_map
    state_dict = model_zoo.load_url(url_map_[model_name])
    if load_fc:
        model.load_state_dict(state_dict)
    else:
        state_dict.pop('_fc.weight')
        state_dict.pop('_fc.bias')
        res = model.load_state_dict(state_dict, strict=False)
        assert set(res.missing_keys) == set(['_fc.weight', '_fc.bias']), 'issue loading pretrained weights'
    print('Loaded pretrained weights for {}'.format(model_name))```

and delete the previous code, it should work
Cyril9227 commented 4 years ago

See my pull request, this change works as expected

cristyioan2000 commented 4 years ago

Yes, it works now, thank you!