Page 1 of 1

Finetuning lowers values

Posted: Sun Jun 05, 2016 10:49 pm
by Lugarimo
My initial noise sample was too dark so I looked around the video for a lighter spot and I found one. When I selected the area, yellow boxes lit up which I understand means that this is a good area to pick. When I press fine-tune the values in channel components, frequency components etc. go up so this is good. However, when I press autocomplete the values drop even lower than the initial noise sample.
I started out with average value of 6.5, finetuning increased to 7 and then autocomplete reduced to 6.

Why is this? Should I do autocomplete at all?
In my 7 years of experience using NV, finetuning usually makes things worse and getting a great initial sample is key or photoshopping it to perfection if I can't.

Posted: Mon Jun 06, 2016 7:45 am
by NVTeam
If the initial noise sample was too dark, then it is perhaps better to use Auto Profile on a lighter area instead of just using Fine-Tune on that lighter area.

Perhaps then, the problem you are describing will not arise at all.

Thank you,
Vlad

Posted: Tue Jun 07, 2016 3:14 am
by Lugarimo
The auto profiler only selects a 64x64 area and it gets a lower value than the one I found, not to mention that the lighter parts have somewhat weaker noise than the darker area.

Posted: Tue Jun 07, 2016 7:38 am
by NVTeam
You can manually select an area and click Auto Profile. This way the profile is based on your manual selection.

Vlad

Posted: Wed Jun 08, 2016 4:41 am
by Lugarimo
But I can't auto profile twice, I can only do that with the initial selection but I need two because the first one is too dark while the second one is missing the darks.

Posted: Wed Jun 08, 2016 9:23 am
by NVTeam
Please Auto Profile using the second one and then Fine-Tune (if necessary) using the darker area.

Vlad

Posted: Thu Jun 09, 2016 12:57 am
by Lugarimo
The darker area contains sharper noise and I can't get as high values with the light area.

https://www.sendspace.com/filegroup/Cl1 ... 8tbhy8AIhF

Posted: Thu Jun 09, 2016 1:11 am
by Tim
The whole light sample area you provided should be used neither for profiling nor for fine-tuning as it contains low-frequency details (gradient). The darker area looks sufficiently noisy and uniform. In fact, noise level is higher for the dark area as well as the profile quality.

Perhaps we would be able to give more advice if you upload the whole frame or send it to us via email to support at neatvideo dot com.

Thank you,
Tim

Posted: Thu Jun 09, 2016 3:56 am
by Lugarimo
Fine, I equalized the frame. This better? http://i66.tinypic.com/2195qvo.png
The darker area looks sufficiently noisy and uniform. In fact, noise level is higher for the dark area as well as the profile quality.
Exactly what I'm saying but then I'm missing half the spectrum. This contradicts the advice Vlad gave me. He told me to focus on the lighter profile.

Posted: Thu Jun 09, 2016 4:14 am
by Tim
It is difficult to give advice without seeing the actual image. Without seeing the image (or the frame from video), we can only give some generic tips. In general, you should be looking for a uniform area (in all frequencies -- high, medium, low), without details, with the strongest noise. Often this area is not too bright or not too dark, but that's not always the case. Generally, it's nothing wrong in using dark areas for profiling as long as the areas are uniform (no details even if you increase brightness) and the noise level there is strong. The area should be as large as possible, but it needs to be uniform in the first place. Please refer to user guide for more details.

The sample of the light area you showed earlier is not uniform. It contains noticeable gradient and therefore should not be used for profiling. Vlad assumed that both areas (dark and light) are uniform and based his advice on this assumption.

If you can share a full frame from your video with us, we could do some experiments and provide more specific advice on profiling and fine-tuning.

Posted: Thu Jun 09, 2016 5:35 am
by Lugarimo
I've made it uniform, see the previous image I uploaded. I know the basics of Neat Video, I've been using it 3 years before your join date. :P

What puzzles me is why this proper noise profile fails to properly remove the noise yet the less accurate lighter profile does the job a lot better. This has been a problem ever since NV4 came out. The numbers are misleading, good profiles are reported as being substandard while crappy profiles are rated as ideal.
Not only can I not rely on the numbers NV gives me, I can't rely on my own eyes either. How except by coincidence would I know that choosing a noise profile from an area that looks nothing like this frame would be the most ideal for killing its noise and not, you know, the actual region from that same frame?
This is frustrating.

Here is the requested screenshot.
http://pasteboard.co/1xX6nLxX.png

This is the only large uniform area I could find. The largest, best lighter area I could find was what I posted in the last post but I did have to photoshop it to be uniform.