Hi,author！I really appreciate your work, and I have been using your code to do some experiment recently. There is a little question about your code in IBCLN_model.py.
In the clip below , you use a if to filter the reflection data which may have large blank area , I can't figure out why you did this. Were there any bad cases occuring during your experiment under this condition?
if torch.mean(T) * 1 / 2 > torch.mean(R):
self.trainFlag = False
returnI tried to use some images with large blank area as R to synthesis I in my training procedure, so I removed this condition. The model had great performance on training data, but worked terrible on test data. Is that any thing to do with the condition?
@JHL-HUST @yangyixiao looking forward to your reply!
Hi,author！I really appreciate your work, and I have been using your code to do some experiment recently. There is a little question about your code in IBCLN_model.py.
In the clip below , you use a
ifto filter the reflection data which may have large blank area , I can't figure out why you did this. Were there any bad cases occuring during your experiment under this condition?I tried to use some images with large blank area as
Rto synthesisIin my training procedure, so I removed this condition. The model had great performance on training data, but worked terrible on test data. Is that any thing to do with the condition?@JHL-HUST @yangyixiao looking forward to your reply!