Closed ph0316 closed 2 years ago
Our model can directly use raw low-light images as input. Proper hand-crafted features as input will further benefit the robustness.
You can manually modify the input $z$ in the code to get different $x_h$
I can't train when I'm using ’NoEncoder‘?Why is that?
That encoder is just for ablation and is out of maintenance.
Maybe you could try to modify the following code from
elif opt['cond_encoder'] == 'NoEncoder':
self.RRDB = None # NoEncoder(in_nc, out_nc, nf, nb, gc, scale, opt)
to
elif opt['cond_encoder'] == 'NoEncoder':
self.RRDB = NoEncoder(in_nc, out_nc, nf, nb, gc, scale, opt)
I am not sure whether this works. If you want to try "NoEncoder", the easiest way is to make the current encoder just output 0.
That encoder is just for ablation and is out of maintenance.
Maybe you could try to modify the following code from
elif opt['cond_encoder'] == 'NoEncoder': self.RRDB = None # NoEncoder(in_nc, out_nc, nf, nb, gc, scale, opt)
to
elif opt['cond_encoder'] == 'NoEncoder': self.RRDB = NoEncoder(in_nc, out_nc, nf, nb, gc, scale, opt)
I have tried to do this, but the picture is black.
You could refer to this issue #18 . It's very difficult to only train the invertible network without conditional features.
Thank you very much!
Hello, I'd like to ask you two questions. The first: why should the input low light image be enhanced by histogram first? Can the low light image be directly sent to the network? Second: you said that you can learn one-to-many using normalizing flow. Where is it?