Use the pre-trained global model weights in Prostate Segmentation example

[Tarun-Khilani]

Hi, I am trying to run the Prostate segmentation example - FedAvg Job but I would like to use a Pre-trained model. I have tried specifying the path in the "config_fed_server.json" by adding the "source_ckpt_file_full_name" in the args of the "PTFileModelPersistor" persistor and also tried to specify it using the "environments.json" file created inside the app folder in the job which contains -
{
"APP_CKPT_DIR": "model",
"APP_CKPT": "pretrained_model.pt"
}
It is still not using the pre-trained model and gives ExecutionException in the case of source_ckpt.

[chesterxgchen]

@ZiyueXu77 can you help when you are free?

[ZiyueXu77]

Hi @Tarun-Khilani , the "environment.json" is more for changing the default saving path of model persistence to a customized one. For starting from a pre-trained model, you can use "source_ckpt_file_full_name", add it to the persistor specs in config_fed_server.json. For example:

{
"id": "persistor",
"name": "PTFileModelPersistor",
"args": {
"model": {
"path": "monai.networks.nets.segresnet.SegResNet",
"args": {
"blocks_down": [
1,
2,
2,
4
],
"blocks_up": [
1,
1,
1
],
"init_filters": 16,
"in_channels": 4,
"out_channels": 3,
"dropout_prob": 0.2
}
},
"source_ckpt_file_full_name": "/media/Experiment/NVFlare/workspace/init_model/init_global_model.pt"
}
}