Page 1 of 1

Jesus! Render is slow

Posted: Fri Jun 25, 2010 4:41 am
by percoplus
My first time using this plug-in (Pro version). The demo was very impressive and quick but I used it on a very short clip and it cleaned up only a portion of the frame. I am now cleaning a 10-minute clip of 1920x1080p Premiere Pro CS4 timeline. mpeg 2 I-Frame footage. It has been rendering for 2 hours. Estimated time left fluctuates and has been 31 minutes for the past hour. Is that normal render time? I have a reasonably fast computer core2 Quad CPU 2.66 GHz, 8 GB RAM 64 bit Windows Vista.

If this is normal render time then the plug-in may not be practical for me to use. What workflow should I use to minimize time?

Posted: Fri Jun 25, 2010 9:38 am
by NVTeam
If the render time is 2.5 hours then the the actual render speed for that clip is about 2 fps. This value looks normal for that frame size (1920x1080) and that hardware.

You can somewhat speed up processing by disabling some of the filter options:
- disable very low freq, high quality options;
- set some of the noise reduction amounts in some or all of the channels (luminance / chrominance or Y/Cr/Cb) and/or frequency ranges to 0%;
- reduce temporal filter radius;
- disable adaptive filtration option.

All these measures can make filtration somewhat faster but at expense of reduced filtration quality. Generally, noise reduction is a tradeoff between quality and processing speed. When you need the top quality, it makes sense to give it some time to process with top settings. When that is not needed, you can use a less accurate and faster filtration.

Hope this helps,
Vlad

Posted: Fri Jun 25, 2010 6:43 pm
by percoplus
So if I have a couple of hours of footage to clean up, it will take the whole weekend. I feel a bit like a sucker having just purchased the plug-in, but I must say it cleans up the video pretty nice.

As for workflow, then, I assume I should de-noise the entire video file and render and export it from Premiere. Import it back in and then edit and apply other effects, color, etc. Or, is it better to de-noise after all other editing is done?

Posted: Fri Jun 25, 2010 7:21 pm
by NVTeam
I think it is better to pre-process imported footage using Neat Video and then edit and compose using a clean version. For the noise filter it is easier to work before any other post-processing. From the workflow standpoint that also helps because noise reduction is done only once and does not have to re-done when you adjust anything in the project.

Vlad

Posted: Fri Jun 25, 2010 9:01 pm
by percoplus
Thanks for the tip. Good product!

Idea: implement an easy "quality slider"

Posted: Wed Nov 20, 2013 3:42 pm
by redrob
I'm a long-time Neat Video client, with the Sony Vegas Pro plug-in -- still love the results, but like the rest wish processing speed could somehow be enhanced. With my HD footage and quad core (can't use GPU .. seems to be a slower option) it slows to between 1 and 2 frames/sec.

What do you think of these ideas:

(1) Examine the clip properties, if possible, to determine if it's interlaced or progressive footage -- on a mixed-media timeline it's a nuisance to have to manually remember which clip is which source type and set it when adding the filter, and also prone to users choosing the wrong option

(2) The NV team has recommended disabling certain noise reduction options to speed up processing (yay!). But this seems somewhat of both an art and a science, and folks like me don't really understand how to disable which options and when it's most appropriate. When simple options are being displayed, perhaps add a slider control that on one end has "highest quality" and the other end "lower quality/faster-renders" -- then, you could behind-the-scenes set the options most appropriate based ont he position of the slider?

(3) My typical projects use 4-7 cameras (3-5 different models), and the amount of noise will vary based on how much gain the operator how to apply when shooting in a changing-light environment. I'd love a "first pass" automatic workflow which looks at the amount of noise in a sample-frame from a clip and guestimates the best-matching noise-profile.

I guess I've never really grasped the noise-profile concept and how it needs to change from event to event. For example, if I own an FX7 and XHA1s -- I can't just create an "FX7-low-gain" and "FX7-high-gain" profile (and same for the other camera) which applies in all circumstances, right? When I've tried that in the past, it sometimes results in too many artifacts (or not enough reduction in some areas) of the image. Any tips, other than going event-by-event?

Re: Idea: implement an easy "quality slider"

Posted: Wed Nov 20, 2013 5:12 pm
by NVTeam
redrob wrote:I'm a long-time Neat Video client, with the Sony Vegas Pro plug-in -- still love the results, but like the rest wish processing speed could somehow be enhanced. With my HD footage and quad core (can't use GPU .. seems to be a slower option) it slows to between 1 and 2 frames/sec.
More or less recent Intel CPUs (like i7 3930K working at 4000 MHz) alone achieve 10 frames/sec in Optimize test.
Fast desktop GPUs can help further as well. The best (fastest) cards for NV currently are:
- NVidia GTX 580/680/590/690/780/780Ti/Titan;
- AMD Radeon HD 7950/7970/7990;
- AMD Radeon R9 270X/280X/290X.
You may want to consider upgrading the hardware if that is a possibility at this point.

redrob wrote:(1) Examine the clip properties, if possible, to determine if it's interlaced or progressive footage -- on a mixed-media timeline it's a nuisance to have to manually remember which clip is which source type and set it when adding the filter, and also prone to users choosing the wrong option
I agree with you, it is a nuisance. It is a pity Vegas doesn't tell the plug-in what kind of footage it is, so the plug-in has to ask the user. There is no other way to correctly process the video data.
Other hosts like Adobe Premiere or Apple Final Cut do tell plug-ins about the scan type, so the plug-in knows how to deal with the frames/fields it receives from the host. If Vegas did the same then this part of the setup could be automated just like it is now in those other hosts/plug-ins.

redrob wrote:(2) The NV team has recommended disabling certain noise reduction options to speed up processing (yay!). But this seems somewhat of both an art and a science, and folks like me don't really understand how to disable which options and when it's most appropriate. When simple options are being displayed, perhaps add a slider control that on one end has "highest quality" and the other end "lower quality/faster-renders" -- then, you could behind-the-scenes set the options most appropriate based ont he position of the slider?
That is a good idea. We will consider how this could be done in the future versions. Thank you.

redrob wrote:(3) My typical projects use 4-7 cameras (3-5 different models), and the amount of noise will vary based on how much gain the operator how to apply when shooting in a changing-light environment. I'd love a "first pass" automatic workflow which looks at the amount of noise in a sample-frame from a clip and guestimates the best-matching noise-profile.
If it looks at the noise in a sample frame and it can be sure that what it looks at is actually noise then most of the hard work is already done and there is not much need for the best-matching noise profile at that point. NV could simply use the profile built from the current frame during that initial analysis. Perhaps that newly built profile would be less accurate than a carefully developed profile built in advance but then such kind of matching is never perfect and the errors of a seemingly-matching profile might be even higher than errors of that imperfect newly built one.. It is not really that simple to make it work any more reliable than the way you do it now. Consider your current setting up workflow:
- you open a frame from the clip;
- you click Auto Profile;
- you visually control the area selected by Auto Profile (to make sure it doesn't select something unsuitable like an area with details or a totally blank area without any noise at all);
- you then check preview in the Noise Filter Settings tab to make sure the results are reasonable;
- and then you apply.

When you use this workflow, you know that the profile is good because (1) you built it using this very clip and (2) you saw the actual results in preview.

In contrast when you use any kind of pre-built profile created using another clip, you somehow automatically load that pre-built profile but you do not control all elements above, so in the end you can receive all kind of results.

Therefore, after such automated loading of some pre-built profiles into all instances of the filter in different clips, you still have to visually inspect them all to be sure in the results you are getting. Will this eventually save any time as compared with the regular workflow above?

redrob wrote:I guess I've never really grasped the noise-profile concept and how it needs to change from event to event. For example, if I own an FX7 and XHA1s -- I can't just create an "FX7-low-gain" and "FX7-high-gain" profile (and same for the other camera) which applies in all circumstances, right? When I've tried that in the past, it sometimes results in too many artifacts (or not enough reduction in some areas) of the image.
That illustrates my point in comments to (3) above. If the noise was always the same for FX7 working in "low-gain" mode then just one "FX7-low-gain" profile would be sufficient to cover all clips created in that mode. But is it really the same? Perhaps not, and those bad results you are describing is an indication of that. Perhaps there are some parameters that still change and affect the actual noise levels. Which is why it is necessary to either build a fresh new profile for each clip or to adjust each loaded profile to the current clip in some way.
redrob wrote:Any tips, other than going event-by-event?
Perhaps the only suggestion would be the following. Try to develop a set of profiles for the most common camera modes that you use and develop them using shots of the Calibration Target.
Then, when it comes to using those profiles, not just load one of them (you still have to manually select the right one of course) but then additionally use the Auto Fine-Tune function in Neat Video. That function adjusts the current profile (and that would be the profile you have just loaded from your pre-built set of profiles) to the current clip, which you have now open in Neat Video. This way, a kind of generic profile would be adapted to that specific clip and will be more accurate and more fitting the noise of that clip.

Hope this helps,
Vlad

Re: Jesus! Render is slow

Posted: Sun Dec 15, 2013 3:57 pm
by rimage
I don't know if this has been addressed but if you have a newer Nvidia GPU you can utilize the unused memory to help speed up processing. I'm currently running a GTX 460 1gb ram. Apparently it doesn't work with AMD GPU's.

Posted: Mon Mar 28, 2016 1:56 am
by BruceUSA
Dear Bruce Phung,

Thank you for the message.

> While back when I trial out Neat video V4 and it was
> impressive fast and I bought it. For the first time I use
> Neat video to reduce noise for a big video project I have.
> Neat video start out very fast and then tapper out to 1-2
> frame. My system is a Windows 7 Pro 64 bit. Intel 6 cores
> processor i7 4930K running @ 4.4ghz custom watercooled. 64GB
> of ram DDR3 1866Mhz. I am running AMD R9 290X also custom
> watercooled. I have been running AMD Driver 15.11 and then
> I upgraded to the lastest AMD Driver 15.12. 15.12 is
> slightly better. But over all neat video render is painfully
> slow. I am rendering a 55 minutes project running at 2 frame
> a time. Any idea how can I improve neat video rendering?

Did you run Optimize Settings tool in Neat Video Preferences?

Please send us the results from that tool for analysis.

Thank you,
Vlad



Hello,

Thank you for your reply. Yes I did Optimize Settings and I accepted the
combination. When I start to render, it starts fast and slowly tapper out
to 1-2 frame. I am still rendering the project. I can not not send you
the result tools for analysis. But When I am finish the project, I will
send you the results tools. Hopefully you can help me figure it out. After
all my system is powerful enough for this type work.


Bruce


I am rendering 10hrs already and is only 30% done. Neat video telling me
there is still 23hrs to go. Unbelieveable......


Bruce