Keras Implementation of Unet with EfficientNet as encoder

• Unet with EfficientNet encoder
  • EfficientNet-B0
  • EfficientNet-B1
  • EfficientNet-B2
  • EfficientNet-B3
  • EfficientNet-B4
  • EfficientNet-B5
  • EfficientNet-B6
  • EfficientNet-B7

1. tensorflow >= 1.13.1
2. Keras >= 2.2.4 (It will automatically be installed when you install efficientunet)

Install efficientunet:

pip install efficientunet

1. This library assumes channels_last !

2. You cannot specify None for input_shape, since the input_shape is heavily used in the code for inferring the architecture. (The EfficientUnets are constructed dynamically)

3. Since you cannot use None for input_shape, the image size for training process and for inference process have to be the same.
   If you do need to use a different image size for inference, a feasible solution is:

   1. Save the weights of your well-trained model
   2. Create a new model with the desired input shape
   3. Load the weights of your well-trained model into this newly created model

4. Due to some rounding problem in the decoder path (not a bug, this is a feature 😏), the input shape should be divisible by 32.
   e.g. 224x224 is a suitable size for input images, but 225x225 is not.

• If you are unable to load the model from the saved HDF5 file, please refer to this issue.
• Especially this comment can be used as a workaround.

0. Some code snippets of EfficientNet are directly borrowed from this repo.
1. The links of pretrained weights are borrowed from this repo.