A question about network structure

[xiongchuanchao]

[xiongchuanchao]

Hello, the + operation in the upper right corner of the network structure, the noisy input add the result after network processing is the denoised output. I can't understand this all the time. Why add the noisy input here. the result after network processing is not the denoised output? @Sumching ChengShen

[FunkyKoki]

This is residual connection, and the whole network is used to estimate the noise in this setting.

[FunkyKoki]

However, I don't know whether you have checked the network architecture. It seems that the author did not use this residual connection in their code.

[xiongchuanchao]

This is residual connection, and the whole network is used to estimate the noise in this setting. but there is no activation function after connection, residual connection is shown in the picture?i am sorry that my deep learning knowledge is poor.

[Sumching]

@xiongchuanchao After DnCNN, recent CNN denoisers prefer to output the residual image and then add it to the input to get the final image. @FunkyKoki you can find it in train.py

[FunkyKoki]

Alright, thank you, I've seen it.

[llz20180725]

Have you ever had a process get stuck？ at model clock