Federated learning using the Digital Slide Archive and HistomicsTK-deeplab
Federated workers can be setup by intalling the Digital Slide Archive
This code uses the Histo-Cloud plugin to train segmentation networks via the DSA instances.
The train_network plugin is called via the REST API to train a network on each DSA instance before the resulting models are downloaded to the master server and averaged.
The Deeplab ImageNet pretrained model can be downloaded from here. This model should be placed in the same folder as this code prior to training.
The hyper-parameters for training can be adjusted in the federated_training.py code.
An example of how to start the training is provided in run_federated_learning.py