Is there any relation between this 'noisy' normal distribution and the 'noise' in path tracing due to low sampling rate? It seems the 'noise' in path tracing makes the output image noisy but this 'noisy' normal distribution makes the image realistic.
saltyminty
I'd wager that the effect of path tracing noise is a lot larger than the effects from a jaggy NDF. For path tracing noise, the drawn ray will hit the wrong objects and/or take the wrong path altogether, which results in the usage of completely different values in our calculations rather than just small offsets.
jierui-cell
These two graphs remind me of what the professor said at the beginning of the class, "noise is your friend when rendering images." I have seen so many examples of this wisdom throughout the class. Adding noise in a smart way significantly increase the realness of an image or 3D object.
waleedlatif1
@yzliu567 I think this is a really interesting point, considering how undesirable path tracing noise is but ensuring that our NDF has noise and isn't smooth to ensure that we can model objects in the real world that do, indeed, showcase some randomness or variability in the texture and shape and lighting. Upon doing some reading, the noise that results from having too few samples in path tracing and the noise that results from the distribution vary. Intuitively, it seems like the best combination of the two is sampling at an efficient rate for path tracing while making up for that heavy computation by estimating the measurements used to compute the NDF, resulting in the 'good' type of noise.
Is there any relation between this 'noisy' normal distribution and the 'noise' in path tracing due to low sampling rate? It seems the 'noise' in path tracing makes the output image noisy but this 'noisy' normal distribution makes the image realistic.
I'd wager that the effect of path tracing noise is a lot larger than the effects from a jaggy NDF. For path tracing noise, the drawn ray will hit the wrong objects and/or take the wrong path altogether, which results in the usage of completely different values in our calculations rather than just small offsets.
These two graphs remind me of what the professor said at the beginning of the class, "noise is your friend when rendering images." I have seen so many examples of this wisdom throughout the class. Adding noise in a smart way significantly increase the realness of an image or 3D object.
@yzliu567 I think this is a really interesting point, considering how undesirable path tracing noise is but ensuring that our NDF has noise and isn't smooth to ensure that we can model objects in the real world that do, indeed, showcase some randomness or variability in the texture and shape and lighting. Upon doing some reading, the noise that results from having too few samples in path tracing and the noise that results from the distribution vary. Intuitively, it seems like the best combination of the two is sampling at an efficient rate for path tracing while making up for that heavy computation by estimating the measurements used to compute the NDF, resulting in the 'good' type of noise.