
38:26
Just curious, what happens when you feed this network actual pure noise?

50:40
does the low bias deflect mean zero noise?

50:53
*reflect

54:40
Batch Norm should cancel the effect of bias in the network even if bias is present. Why is Batch Norm not able to cancel the bias in DnCNN ?

58:34
Thankyou

01:03:13
How many of the removed constants were non-zero?

01:04:08
thanks

01:18:31
A couple of questions if possible:

01:23:36
sorry in the end

01:36:03
Can you give an intuition for why is the p(y) is getting finer and finer over iterations (blurry to fine)?

01:50:54
sorry, it was accidental touch while gripping my mobile...

02:00:23
What is the chance the interpretation of the cone of the denoiser in 3D would change if you change your objective?

02:00:28
Thank you!

02:00:42
Great Talk, have you looked into other metrics besides SSIM to assess image quality?

02:01:18
yeah

02:09:11
Once we have obtained the prior and from the examples shown seems pretty good would that be affected if the input changes? Think of prior obtained from grey scale inputs and trying to apply to rgb inputs?