cszn / SCUNet

Practical Blind Denoising via Swin-Conv-UNet and Data Synthesis (Machine Intelligence Research 2023)
https://link.springer.com/article/10.1007/s11633-023-1466-0
Apache License 2.0
637 stars 63 forks source link

the issue of train iterations in your paper #21

Open lhehejunl opened 1 year ago

lhehejunl commented 1 year ago

Dear author: In your paper, The learning rate decays by a factor of 0.5 every 200,000 iterations. I want to confirm if it's 200,000 iterations, because in my opinion, 200,000 iterations is a very large number. The learning rate starts from 1e-4 and decays by a factor of 0.5 every 200,000 iterations and finally ends with 3.125e-6. Thus, a total of 100,000,0 iterations are required, right?