Questions about decoder output.

[CaiShilv]

It is a great honor to see your work. I would like to ask why the decoder output is the residual instead of getting the reconstructed image directly.

[mengxzh15]

我觉得吧，当然可以是直接输出了结果了。你自己训练一遍就行。

[xiaogang00]

Yes, you can also train the model to output the image directly. We find that predicting residual can lead to better results, and this is why we adopt residual learning.