About the preprocess of the dataset alignment

[camellia120]

Hi aiff22,
when i used the network in my task , it seems that the network not only learns the changes of content and texture and etc, but also learns the deformation between GT and Input images.
The preprocessing of the dataset is also basically similar to the preprocessing you listed here(aiff22/DPED#7). So, is it possibile the network learns the deformation or the differences of the preprocess casue the results?
Finally, can you provide the code of preprocessing data?

[aiff22]

Hi @camellia120,

You are right, the network indeed tries to learn the mismatch between the source and the target images. However, this mismatch is almost random and it comes not from the pre-processing step, but from the nature of the data: different optical systems and camera sensors are causing different distortions and aberration on the resulting photos. Unfortunately, this cannot be fixed with any existing software.

[camellia120]

Hi @aiff22,

Thx for your reply.
The mismatch caused by optical systems and camera sensors is almost random: it can be expected when it has enough data in training. Beacuse the netwrok learns the balances of the mismatch of all sample pairs. And the mismatch in different sample paris is indeed random. In my experiments, the mismatch is not almost random when in testing. Maybe the network is overfiting because of lack of enough data. But i doubt it also may causes by my preprocess. I would appreciate it if u can send the script of preprocess to my email: [email protected].

[7Yearjksahdjk]

Hello
The author doesn't seem to have seen my source file request, I would like to get the PyNET source code from you.
I hope to get your support!