search

jiangsutx / SRN-Deblur

Repository for Scale-recurrent Network for Deep Image Deblurring
http://www.xtao.website/projects/srndeblur/srndeblur_cvpr18.pdf
MIT License
710 stars 184 forks source link

Weight Decay #11

Closed ducheng678 closed 5 years ago

ducheng678 commented 5 years ago

Do you imply weight decay when training SRN? I do not find weight decay in the code.

jiangsutx commented 5 years ago

No. We did not use that.

ducheng678 commented 5 years ago

Did you have any experiments of using weight decay in SRN? If you have, how did the model perform?

jiangsutx commented 5 years ago

No. We didn't. Weight decay is usually used when over-fitting. We did not observe such phenomenon, so no such experiments have been conducted.

ducheng678 commented 5 years ago

ok, thx.

ducheng678 commented 5 years ago

Have you conducted experiments about u-net structure? I didn`t find any comparision about processing results of u-net in your paper.

jiangsutx commented 5 years ago

We have a SC baseline model for using single scale in the paper. That is a U-shape net with ResBlocks. Original U-net performs poor for this task.