Confusion about Cascade Training

[pzhhhhh2263]

hello,"Note that the 3D full resolution U-Net of the cascade requires the five folds of the low resolution U-Net to be completed!" If I use fold all, can I start training the 3D full resolution U-Net once I finish the fold all low resolution U-Net?

[seziegler]

Hi @pzhhhhh2263 , no, fold all means that the model is training on all the available training data (without cross validation splits). For the cascade you need the low resolution models trained on the folds. If you don't want to use cascade you can also simply only train the full resolution models.

[pzhhhhh2263]

Does this mean that if I want to use cascade training, I must use five-fold cross-validation?But after receiving your reply, I only trained for one round on 3d_lowres, and it still produced results in a folder named predicted_next_stage.

[seziegler]

If you don't want to use cross validation at all you can also just do everything with fold all