How is v3 better quality?

general questions about Neat Video
Post Reply
Lugarimo
Posts: 114
Joined: Mon Feb 09, 2009 2:51 pm

How is v3 better quality?

Post by Lugarimo »

It is described as producing better quality output from noise than before. Can this be explained better?

The sample #2 video of that lady in a chair I cant tell difference from before. In fact, I think it even looked better before.

If v3 is really better quality, is it better quality for heavy noise or light noise?
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

All examples have been redone and also recompressed to allow online playback in browser, so it is possible that something may look different since the codecs are different and compression is lossy.

Regarding the quality improvements in general, they are in two areas where they should be visible:
- the temporal filter is now able to distinguish more actual details and better preserve them; for example, in the current testkit clip, you can notice that in the areas with wires, features of the cars, road marking, those are better preserved now;
- the noise analyzer now more accurately measures the noise properties, so the resulting profile is more precise, which generally helps the filter to do accurate noise reduction; this also helps the Adaptive Filtration feature, which adjusts the noise filter to the noise changing along the timeline; these adjustments are more accurate now; you will see this when working with a clip that contains several continuosly changing scenes with different light conditions for example.

Please test the Demo version. If you find that v3 quality improvements will not be significant in your projects, then you can always continue to use the v2. However, even in that case, you can still benefit from faster processing in v3.

Vlad
Lugarimo
Posts: 114
Joined: Mon Feb 09, 2009 2:51 pm

Post by Lugarimo »

Cool, so let me see if I understand since I always did this even with v2 anyway. I would always play the whole clip (even if its hours long) and look for the worst noise scene and profile that area and apply it to the whole video even if the low-noise parts would lose some details because of it.

Are you saying that now I wont have to worry about it since the adaptive filtration is much better?

Also, can you tell me how you guys cleaned up sample #2, I still cant seem to be able to get as good quality as you guys did.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

Lugarimo wrote:Cool, so let me see if I understand since I always did this even with v2 anyway. I would always play the whole clip (even if its hours long) and look for the worst noise scene and profile that area and apply it to the whole video even if the low-noise parts would lose some details because of it.

Are you saying that now I wont have to worry about it since the adaptive filtration is much better?
I would say that if you use the same method with v3, it will lose less details because Adaptive Filtration will better adjust the filter along the clip.
Lugarimo wrote:Also, can you tell me how you guys cleaned up sample #2, I still cant seem to be able to get as good quality as you guys did.
There were no special tricks, just building a good profile and rendering. The same method as before.
Lugarimo
Posts: 114
Joined: Mon Feb 09, 2009 2:51 pm

Post by Lugarimo »

How do you build a good profile with such a tiny video?
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

You can use a small area for profiling too. In that clip, it is possible to find a 64x64 area.

Generally, it makes little sense to try to exactly reproduce our result using the clip from the online example because it has already been compressed for online replay and thus already contains changes introduced by the lossy compression. It makes sense to work with the original clip instead. If you want to do that please use the clip from our testkit. Run it with v2 and v3 and compare two frames by flipping between them in an editor or image viewer, you will see the difference.

Thank you,
Vlad
Lugarimo
Posts: 114
Joined: Mon Feb 09, 2009 2:51 pm

Post by Lugarimo »

Testkit doesnt contain the sample 2 video.

I know I can select a 64x64 area but I can't get any % value higher than 56%, what value were you able to get?

I do still have copies of your old and new processed versions of the sample 2 clip and I compared them side by side, but they are too different to reasonably compare. Old one is more yellow and it looks like you guys used 100% Y removal while only maybe 60% on the new one. In some areas the new one looks better but it also has more wobbling. The old one is blurry on scenes where she moves her head but the new one doesnt have this problem.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

It was 37%. The goal was good filter results, not high quality value of the profile per se. The quality value does not necessarily need to be high, the profile just needs to be built correctly.

Vlad
Lugarimo
Posts: 114
Joined: Mon Feb 09, 2009 2:51 pm

Post by Lugarimo »

I still can't get as good quality as you guys. Did you use any adjustments like more temporal filtering, more high frequency noise filtering and such?
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

We only enabled sharpening there, everything else was by default. You will probably not get exactly the same result when using the clip compressed for the web. I recommend to use your own clips for testing. After all, that is what really counts.

Vlad
Post Reply