This work shines a light on many interesting things from the right training strategy to the regularizations that hinder network performance. However, it would be wonderful if you find time to address the following questions:
How many times experiments were run for Tables 3-5? Can you provide statistics of experimental results (e.g. mean, variance)?
What was the reason for using self-ensemble?
Do you have any plans to extend the results to other image restoration methods (e.g. deblur, denoise)?
Dear authors,
This work shines a light on many interesting things from the right training strategy to the regularizations that hinder network performance. However, it would be wonderful if you find time to address the following questions:
Thank you!