search

wbhu / DnCNN-tensorflow

:octocat::octocat:A tensorflow implement of the paper "Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising"
GNU General Public License v3.0
363 stars 150 forks source link

Without the BN layer can performance well #29

Open chaoyueziji opened 6 years ago

chaoyueziji commented 6 years ago

Hello，When I was training the network, I did not add the BN layer, and the other structures were the same as the original. The training set used is only BSD40, and patch size is 40*40. In the test set12 on the PSNR is higher than the original, and the results from the loss convergence is also faster convergence than the BN layer.

wbhu commented 6 years ago

You could put the quantitative results and the trained model for evaluation.

xiaojuan123 commented 5 years ago

but my network does not converge when I remove BN，what's your resolution?please

chaoyueziji commented 5 years ago

but my network does not converge when I remove BN，what's your resolution?please

Hmm, my network structure is Conv + relu + bn, so when I remove BN, I find it convergent and has similar performance. So at first I thought BN was an unnecessary module. My experiment result is to remove BN module and optimize it by Adam. When the noise level is 25, the result can reach 29.15.