InnerEye-DeepLearning
InnerEye-DeepLearning copied to clipboard
Create a segmentation model diagnostic notebook
- Image sizes
- effect of photometric normalization
- patch sampling thumbnails
Add average Dice to results notebook
Can we maybe even add GPU utilization stats, so that people can increase crop size if there is space?
We are not always evaluating the model on the full training set, but maybe we can add performance per crop from the last training epochs? Can that give any insights into the quality of the training data?