search

kumar-shridhar / PyTorch-BayesianCNN

Bayesian Convolutional Neural Network with Variational Inference based on Bayes by Backprop in PyTorch.
MIT License
1.4k stars 323 forks source link

Image noise problem #10

Closed GGyan closed 5 years ago

GGyan commented 5 years ago

when I use the Bayesian convolution neural network to process image problems, the output image has noise. Do you know why this is?

kumar-shridhar commented 5 years ago

Hi @GGyan, Can you please provide more information? Like what file you ran on what dataset and where was the noise?

GGyan commented 5 years ago

I am very happy with your reply,When I use Bayesian convolution neural networks, I define a convolution layer and then input the dataset for training. After finishing the training, I input a picture and test it. The result showed that the picture produced noise. I don't know what caused the noise and hope to remove the noise. My dataset: https://sites.google.com/site/boyilics/website-builder/project-page. Because I am studying image defogging, my dataset is divided into two parts: artificial synthetic haze image and the original no-haze image. The results of my test are shown below:

On 5/15/2019 15:36，Kumar Shridharnotifications@github.com wrote：

Hi @GGyan, Can you please provide more information? Like what file you ran on what dataset and where was the noise?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

GGyan commented 5 years ago

man noiseman

kumar-shridhar commented 5 years ago

Did you use the super resolution code for this? I think the result is interesting and maybe due to the latest changes in the Bayesian distribution layer, the noise is added due to an inefficient sampling method use. The results should be fixed with the code update that I will push soon (in 10 days). Thank you for addressing the issue.

GGyan commented 5 years ago

Thank you for your timely answer.

kumar-shridhar commented 5 years ago

Hi @GGyan, can you try to run your code now and see if anything changes.