EfficientNet (PyTorch)

My implementation of Efficient-Net in nn.Sequence manner (i.e. without nn.Module) for feature extraction layers. This approach is more convinient for applying Class Activation Mapping or working with fast.ai library.

About EfficientNet

To load pre-trained model simply run

import efficientnet
model = efficientnet.efficientnet(net="B4", pretrained=True)

where B4 could be replaced with any model scale from B0 to B7. Weights will be downloaded automatically.

For features extraction simply run

import efficientnet
image = torch.randn(1, 3, 300, 300)
model = efficientnet.efficientnet(net="B4", pretrained=True)
features = model.features(image)

In same way you can get output from any layer.

Weights were copied from here and adopted for my implementation.