Best Validation Loss

The lowest validation loss achieved during training — indicates the best model checkpoint.

During training, the model checkpoint with the lowest validation loss is automatically saved. This "best" model is used for producing your final denoised images.

Why Best Val Loss Matters

• Validation loss measures performance on unseen data

• The epoch with lowest validation loss typically produces the best denoising quality

• Later epochs might overfit, making the early "best" checkpoint superior

What the number means

• Lower is better

• Typical values depend on your data and noise level

• Compare relative values within your training run, not across different datasets

How it's used:

• The model checkpoint at this loss value is saved

• This checkpoint denoises your images when training completes

• If you download the model, you get this best checkpoint