UNET layers frozen - Githubissues

qubvel / segmentation_models

Segmentation models with pretrained backbones. Keras and TensorFlow Keras.

MIT License

4.75k stars 1.03k forks source link

UNET layers frozen #393

Open tlc10 opened 4 years ago

tlc10 commented 4 years ago

Hi,

I'm currently using your API for a personnal project. While using a UNET model with "seresnet50" backbone and "encoder_freeze=False", I would like to know what does it means concretely ? How many layers of the model are frozen ? Are the weights adjusted during training? etc..

Thank you for your inquiries!

qubvel commented 4 years ago

encoder_freeze = True - encoder is freezed (e.g. resnet18 or densnet121) In another case all layers are trainable.