The denoising effect is abnormal

[sunyclj]

when run ‘python main_ViDeNN.py --use_gpu=1 --checkpoint_dir=ckpt_videnn --save_dir='./test_data/result/frame1_videnn' --test_dir='./test_data/test/frame1'’，the image after denoising is abnormal. original image denoised iamge excuse me, what is the reason?

[MexWayne]

when run ‘python main_ViDeNN.py --use_gpu=1 --checkpoint_dir=ckpt_videnn --save_dir='./test_data/result/frame1_videnn' --test_dir='./test_data/test/frame1'’，the image after denoising is abnormal. original image denoised iamge excuse me, what is the reason?

还好看到你的结果，不然我也就花时间复现效果了

[MexWayne]

I‘ve got even worse results than you，I will give up this work （T_T）

[MexWayne]

when run ‘python main_ViDeNN.py --use_gpu=1 --checkpoint_dir=ckpt_videnn --save_dir='./test_data/result/frame1_videnn' --test_dir='./test_data/test/frame1'’，the image after denoising is abnormal. original image denoised iamge excuse me, what is the reason?

I've got the normal results by input the dark-light case image

[ajeema]

The denoising is excellent, however I am getting the same issues as above. How can I resolve this?

[clausmichele]

This issue has been reported already several times: https://github.com/clausmichele/ViDeNN/issues/1 https://github.com/clausmichele/ViDeNN/issues/7 https://github.com/clausmichele/ViDeNN/issues/8 https://github.com/clausmichele/ViDeNN/issues/20

A possible solution is to use a different checkpoint as explained here: https://github.com/clausmichele/ViDeNN/issues/1#issuecomment-442762424

Another solution would be to retrain the network from scratch.

[ajeema]

The other checkpoint was better but still had artifacts. I'm retraining the network now. If It produces better results I can post the checkpoint if you like.

[ajeema]

retrained and get same results. I wonder if it needs different noise generation? What do you suggest?