Table 2 Hyper-parameters for Model Training
From: Generalizable deep learning framework for 3D medical image segmentation using limited training data
Hyper-parameter | Value |
|---|---|
Patch size | 128a |
Learning rate | 0.001a |
Learning rate drop period | 4 - 6 |
Learning rate drop factor | 0.8 - 0.9 |
Encoder depth | 4 |
Optimization method | Adam |
L2 regularization strength | 1e-4 |
Gradient Threshold Method | L2 norm |
Gradient decay factor | 0.9 |
Squared gradient decay factor | 0.9990 |
Epochs | 100 |
Minimum batch size | 30 - 100 |
Training / testing type | Hold-out a |