NeRD-Neural-Reflectance-Decomposition icon indicating copy to clipboard operation
NeRD-Neural-Reflectance-Decomposition copied to clipboard

Published 20 hours ago • verified publisher icon cgtuebingen

Blank Results

Open Woolseyyy opened this issue 2 years ago • 3 comments

I am trying to follow and reproduce your work but meet some problem. I tried the script in README but find the rgb output is all white. This is my command: python train_nerd.py --datadir /root/Data/synthetic/Chair/ --basedir ~/Experiments --expname 03_nerd_official --gpu 0,1,2,3 --config configs/nerd/blender.txt

I checked that the ground truth image can be correctlly read. When debugging, I find that 'target' and 'target_mask' are all correct but payload['rgb'] always output 1.0 after training. Any ideas? Thanks a lot!

Woolseyyy avatar May 06 '22 12:05 Woolseyyy

I'm running into the same issue with my custom data --- in some cases the predicted alphas in fine_result['acc_alpha'] would become all zeros. Any help would be appreciated :)

za-cheng avatar May 19 '22 16:05 za-cheng

I'm running into the same issue with my custom data --- in some cases the predicted alphas in fine_result['acc_alpha'] would become all zeros. Any help would be appreciated :)

I still cannot solve this issue after debugging and give up. It seems that training nerd is unstable. Have a try with Neural-PIL. I can stablely reproduce the results of Neural-PIL.

Woolseyyy avatar May 20 '22 03:05 Woolseyyy

I'm running into the same issue with my custom data --- in some cases the predicted alphas in fine_result['acc_alpha'] would become all zeros. Any help would be appreciated :)

I still cannot solve this issue after debugging and give up. It seems that training nerd is unstable. Have a try with Neural-PIL. I can stablely reproduce the results of Neural-PIL.

Hi, I am using Neural-PIL for relighting. What should I do? The README.md section doesn't seem to mention how to relight.

MoreThanLove avatar Nov 25 '23 03:11 MoreThanLove