New 4.0 neatvideo

general questions about Neat Video
jpsdr
Posts: 221
Joined: Mon Aug 11, 2008 7:33 am

Post by jpsdr »

Do you mean that without slow shutter and without spot removal, you've encounter no issue ?

If i'm asking, it's because for my personnal use, i intend to use it the "same way" than NV3, without these extra features.

Have you made tests of NV4 vs NV3 without using the new features ?
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

Lugarimo wrote:and this time I created the profile with both NV4 and NV3.
The profiles must be built separately in these different versions of NV, that is by definition. I am surprised that was not done before.
Lugarimo wrote:NV4 either removes too much or too little noise.
With default filter settings?
Lugarimo wrote:Having the spot removal and slow shutter on at the same time created random blotches of smudging on some frames and removed parts of legitimate objects. Turning slow shutter off fixed this for the most part but some problems still persisted.
Does the clip require using that option?
Lugarimo wrote:Turning removal down to 100% alleviated most of them but now most spots and scratches are faded rather than removed.
I am wondering if this is a case of some very difficult clip with a combination of problems (repeated frames, irregular noise, dust/scratches) or perhaps it is just a problem of using an inaccurate noise profile. It is impossible to determined that from your words because you just share your overall impressions. It is best to prepare a good test case and send it to support for detailed inspection. Then we will know who is responsible for the poor results - the profile, filter settings, filter itself, or something else.
Lugarimo wrote:On top of all this, NV4 is 3 times slower than NV3 so what's the advantage here?
With default filter settings, NV4 is about the same fast as NV3. I wonder if you switched it to the high quality mode. Lets see the test case, that will be visible there.

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

Post by Lugarimo »

Test case: https://www.sendspace.com/file/f7j7qa
EDIT: I forgot to include the NV 4.0 DNP in that RAR. Here they both are:
https://www.sendspace.com/filegroup/ghl ... jjXBEACbsg

I also included the frame I chose as the best for a noise profile because it contained the sharpest noise with a large area and got the highest score as well.
NVTeam wrote:With default filter settings?
Better than default, I increased the amount of removal because it's a cartoon.
NVTeam wrote:Does the clip require using that option?
Yes sir, the video has duplicate frames (as all cartoons do) and film damage in addition to sharp film grain. I don't need NV for the spots and scratches though, Avisynth filters take care of that just fine. The fact that the spots are repeated on duplicates removes the effectiveness of the filter anyway. The duplicate frames are normally removed first with Avisynth but the sharp noise interferes with the duplicate detection algorithm so it's a hell of a dilemma.
NVTeam wrote:With default filter settings, NV4 is about the same fast as NV3. I wonder if you switched it to the high quality mode. Lets see the test case, that will be visible there.
I use high quality with both NV3 and NV4. NV4 is 3 times slower.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

High quality mode in NV4 is a whole new option not present in NV3. It is really something very different and it does takes longer (1.5-4.0 times longer) by design. If you want NV4 to work comparably to NV3 then please run NV4 in normal quality mode, do not use that high quality mode.

Thank you for the test case, we will download and run it to see what is going on.

Vlad
jpsdr
Posts: 221
Joined: Mon Aug 11, 2008 7:33 am

Post by jpsdr »

NV3 has a "High Quality" and "High Resolution" box check.
So it's expected people will think it's the same thing that "High Quality" mode of NV4.
If not, what are they in NV4 ?

I think Lugarimo you're mistaking about the slow shutter mode. From what i've understood it's for when there is duplicate frames, but also the noise is identical. In anime, of course there is duplicate frames, but the noise is not the same from one frame to another, and so it's not a slow shutter condition.
From what i've understood, slow shutter if when for exemple you may have ([N] = Frame number N, (n) : noise of frame N) something like that :

Code: Select all

[N]+(n), [N+1]+(n+1), [N+1]+(n+1), [N+3]+(n+3), ....
                            |
Duplicate frame and noise here, frame N+2 is a complete duplicate of N+1, even the noise.

But, with anime footage, you have something like that :

Code: Select all

[N]+(n), [N+1]+(n+1), [N+1]+(n+2), [N+3]+(n+3), ....
Original picture is duplicated, but noise added is frame independant. So, i think it's not a slow shutter case.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

jpsdr wrote:NV3 has a "High Quality" and "High Resolution" box check.
So it's expected people will think it's the same thing that "High Quality" mode of NV4.
We removed that old v3 checkbox and the corresponding function when developing v4. That function was rather minor in terms of its impact on speed and quality.

In v4, we have created a whole new mode of filter operation where it works more thoroughly and with higher accuracy in many more components of the filter. Because that affects many more parts of the filter, the control of this new mode is moved to the global filter settings and (2) its influence on the render speed is much more noticable now.

So the name may look similar but under the hood it is a completely new option, it is not just a one-to-one replacement of the old one. This new option is explicitly mentioned in the announcement of v4 as a new function:
"High Quality" mode for better detail preservation and sharper image (at the cost of reduced processing speed: 1.5-4 times slower)
So the speed drop is to be expected as specified in the documentation. If you want to keep the original speed then just use the normal quality mode, which is about equivalent to v3 speed. I hope it is clear now.

Vlad
jpsdr
Posts: 221
Joined: Mon Aug 11, 2008 7:33 am

Post by jpsdr »

Thanks, for the information.

But... i was more interested if you confirmed or denied my understanding of the slow shutter mode.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

Exact copies of the whole frame content are treated automatically, no need to enable the Slow Shutter mode for that. Exact copies are indicated by "Copy" in the thumbnails.

Almost exact copies do require the Slow Shutter mode. Those are indicated by "Rep" in the thumbnails.

The Slow Shutter mode is for situations where the whole frames are almost exact copies of other frames OR only the noise part is duplicated (almost exact copy) from one frame to another. These two cases are shown in two examples here.

Vlad
jpsdr
Posts: 221
Joined: Mon Aug 11, 2008 7:33 am

Post by jpsdr »

So, to be sure i understand properly, for the anime case, in which content are often exact copy, but noise is NOT duplicated, it's not slow shutter mode.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

If the noise does regularly change from frame to frame then you do not need to enable the Slow Shutter mode.

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

Post by Lugarimo »

NVTeam, the speed is not finally the point. If it were to be many times slower, I would expect better results. I'm consistently getting worse results with NV4 despite the slower processing. Either too much noise removed or not enough.
In that case, not enough.
Have you checked out the test case I uploaded?
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

Lugarimo,

My colleague checked the test test and told me that the noise profile does not suit that clip well enough, which is why you get poor results. He prepared a new noise profile/preset, which produces better results. Please try it. There is also a sub-folder in that zip containing an adjusted filter preset that makes your profile work better. It raises some noise levels a bit to compensate the inaccuracy of the profile and adds adaptive to potentially help with changing noise.

Bottom line: the problem is caused by using an inaccurate noise profile (which may be good in itself but it doesn't match the actual noise in the clip then the results are expectedly poor). Try to build a more accurate noise profile. If you see that you lose details, that is a sign of that. If you see that there is not enough noise reduction, it is a sign too.

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

Post by Lugarimo »

Update: That new profile you gave me works a lot better than mine even on the scene with the sharp noise.

In fact, according to the profile info it has more components in all frequencies and channels even though it has a lower rating.

But visually I do NOT see it having more components, especially high-frequency ones. The noise area you profiled is a lot duller than the big area I picked. How the hell is it sharper according to NeatVideo?

This makes no sense. First I can't rely on the % rating and now I can't rely on my eyes? It's like what I'm using is not NeatVideo anymore.

So help me out, how do I judge which noise area is best to profile if everything I knew for the past 6 years is now obsolete?
Last edited by Lugarimo on Sun Aug 16, 2015 4:55 am, edited 1 time in total.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

This particular problem with that clip seems to stem from the fact the noise changes significantly within the video you work with. It is more or less the same in the clip you sent us but perhaps it changes more drastically in other parts of your video. Which means that you either need to somehow apply different processing to different parts with different noise, or to build a single profile based on samples of strongest noise, which may even require manual fine-tuning using different frames, if those samples of strongest noise are not present in the same one frame. Using adaptive processing can be useful too, but starting with the "strongest noise" profile is the best way.

I am not sure how you built the noise profile used in the test case but my collegue built his profile using the clip from the test case itself. He used a simple technique to build his profile: he manually selected an area in the frame where the noise was strongest among 11 frames available in the window (it is easy to compare noise levesl in those frames now in v4), then just clicked Auto Profile.

Your noise profile was probably built using another part of that larger video, which is why it is not very suitable to filter this part presented in the test case.

So again, we used the most simple way of building a noise profile for this clip. The only trick was to select the noisiest sample.

I recommend to try the following scenario in your situation:
1. build a noise profile using the regular methods described in the userguide and video tutorials;
2. check if noise reduction looks insufficient in some scenes (without and then with adaptive processing enabled); if it does look insufficient then additionally manually fine-tune the profile using frames (if necessary, multiple frames) from those scenes);
3. if the above doesn't help still, build a fresh new profile using a completely different primary noise sample area and repeat.

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

Post by Lugarimo »

Vlad, you are not telling me anything new and you're not listening to what I'm saying.
it changes more drastically in other parts of your video.
Drastically is a stretch, it changes but not that much and I already said NV3 didn't have this problem.
to build a single profile based on samples of strongest noise, which may even require manual fine-tuning using different frames, if those samples of strongest noise are not present in the same one frame. Using adaptive processing can be useful too, but starting with the "strongest noise" profile is the best way.
That is exactly what I said I normally do and have been doing for the past 6 years without a problem until now.
I am not sure how you built the noise profile used in the test case
I included the frame where I built the noise profile from in the test case. LOOK at it. The noise is sharper and stronger than the area you or your colleague selected.

Yet somehow... it is a less accurate profile than yours. According to Neat Video's % rating it is less accurate and according to any normal person's eyes it is less accurate yet Neat Video considers this accurate. HOW DO I RECONCILE THIS, SIR?

Do I intentionally pick noise areas that look the least like what I'm trying to remove or what?

Look at cw noise profile frame.png and explain to me exactly what is wrong with this frame that I chose to make a noise profile.
Post Reply