CCM-Pro

Progressive Distillation for Incremental Learning in Corneal Confocal Microscopy Segmentation

The CORN-Pro dataset can be download from Zenodo CORN-Pro

Running Training

1. Adjust the data path to match the required structure:

   • train -> image, label
   • val -> image, label
   • test -> image, label

2. Begin by training the CNs task. Without specifying the model_initialization function, select models model_S and model_T, then run main.py to obtain the optimal segmentation model for CNs.

3. Next, load the optimal weights from the CNs task using the model_initialization function. Then, invoke model_S_adapter and model_T_adapter, and run main.py to derive the optimal segmentation model for LCs.

4. Finally, use the same training setup for the LCs task to obtain the optimal segmentation model for the SCs task.

Testing

Load the optimal weights for the task, and then execute test.py.