I have to say, the Denoise ML is really a "one click" wonder!

@Reyfox Not to me. Denoise ML is extremely resource heavy. Not practical as yet to be used in an everyday editing workflow.

@Steve_Rhoden True, but at least the editing can be done fluently with Denoise ML added but disabled, and when it is time to render enabled, I have written 2 specific scripts just for this purpose one for disable, one for enable:

https://www.vegascreativesoftware.info/us/forum/fix-for-slow-bcc-denoise-ml--143643/#ca898923

I use it when needed in my workflow, enjoy!

Boris FX's official advice is to render the clips first.

That is how I have used it. See the results, disable, and continue to edit. On rendering, turn it back on. Yes, it is resource heavy.

I am curious if anyone has NeatVideo and Denoise ML what are the results and speed.

@Reyfox Neat Video is what i use and it is very, very fast right now...... Back in the early days of say version 1 and 2 it was painstakingly slow, just like how Denoise ML is at the moment. But the developers kept pounding and refining it over the years, and now it is simply perfect, super-fast, efficient and use little to no resources.

So i hope likewise overtime, Boris will keep on refining these machine learning filters to be faster and much more efficient (They say they will, so we will see).... But for now, it is strictly neat video for me.

@Steve_Rhoden thanks for the reply. I have NeatVideo for another editor. I liked how you can let it test your computer to see what combination of the CPU and GPU will give you the best results. And I too found it really fast. But since I shoot almost everything I edit, I don't shoot "in the dark", so had a friend supply me a video clip he had with a lot of noise. I found that DeNoise worked quite well with one click. Of course, timeline playback of the denoised clip was painful. Very much so. Even rendering was like at 2fps on a 1080 25P clip, with mostly the load shifted to the CPU.

But, since I have Continuum 2024 and it's included with it, there is no need now for me to purchase NeatVideo for Vegas.

I am counting on Boris to continue to improve DeNoise.

I found that DeNoise worked quite well with one click. Of course, timeline playback of the denoised clip was painful. Very much so. Even rendering was like at 2fps on a 1080 25P clip, with mostly the load shifted to the CPU.

@Reyfox it cycles between CPU and GPU. With your 16core CPU are you hitting 100% when it cycles to CPU?

I am wondering if there is a limit to CPU use or if it will use all available CPU. for 1080P I got 150W GPU use with compression + noise, and 160W with camera noise alone.

I've just been using BCC+Upres ML and am so impressed with it. BCC+Upres ML provides significantly better results than BCC Uprez (IMO) which I was previously using.

@Dexcon I ran some upscalers at default 2x, I didn't like the results from any of them with my sample. My source was 480P with compression artifacts so output 960P. What you'd normally do is make adjustments on a frame until desired result,,, if there were one. I didn't do that.

@Former user, rendering with AMD VCE, 1080 25P footage, the max the CPU saw was about 49%. It would go up and down, with a low of 7%. The GPU would go between 7% to 4%. I would be curious as to how NeatVideo handles the denoise when rendering.

For me, my few early tests with BCC+Denoise ML was not impressive - nowhere near Neat Video processing.

I have NeatVideo in another editor. The ability to optimize performance with CPU/GPU I think is stellar. But, I rarely shoot scenes where it's required. But I don't have it for Vegas to compare speed and final results.

@Dexcon I can understand computer performance would be different, but how does the final output compare between Neat and Denoise?

@Reyfox  ... From what I've seen so far, BCC+Denoise ML has few adjustments in comparison to Neat DN. I need to test further, but at the moment Neat DN remains my preferred DN VFX. Interestingly, Vegas Pro's native Denoise VFX is more effective than BCC+Denoise ML.

@Dexcon I haven't looked at Vegas denoise. I'll give it a try and see how it goes. A lot of the higher end video editing software all have a denoiser as part of their effects.

@Dexcon, I just did a test with the same clip. The Vegas Denoise used 87% GPU and was fast. Tweaking is a little better. But the results, depending on how much noise, slightly favored Boris in some sections of the clip I was using. Maybe some more tweaking is necessary with Vegas Denoise on my part. But timeline playback definitely favored Vegas.

@Reyfox The denoise has a beauty filter look to it with faces, it's softening and removing features. The upscale on faces gives the plastic skin effect especially on highlights. I only tried on the 480P clip above. It reminds me of early Topaz models, not that latter versions have improved that much with faces, maybe less plastic more cgi.

The ML filters may behave more positively on high quality source faces and non people. It's also why the Resolve enhanced super scale looks so bad, can only feed it high quality camera footage.

Also the upscaler is acting a bit strange, give it 1080p footage in a 2160p project, speed wise it's pretty good, but then make a 2160p video of the 1080p clip without scaling, so black borders on all sides, use upscaler at 200% giving same end result, but (for me) much much slower.

@Former user are you referring to DeNoise ML? I've carefully compared when using it with the original 1080 noisy footage to see if there is any softening on faces, hands, etc., and have not noticed any myself.

I can't answer for 480P footage.

I've found, without changing any of the stock parameters with DeNoise ML, that it works exceedingly well at removing noise in the footage I've tried it on. No, whatever clip it is applied will not play back smoothing at all on the timeline like the Vegas Denoise, which again, works faster utilizing, at least on my computer, the CPU heavily. Nor have I compared the DeNoise ML results against NeatVideo. I'll leave that to someone who has both.

And since it is included with Continuum Complete, I'm quite happy with the results.

I asked Peter McAuley, Senior Product Manager for Continuum some questions about the options within DeNoise ML. Here is his response in italics:

1. What is the difference between 16/32 bit float, and Default?

Default : GPU hardware render

16 bit float : GPU hardware render
Float 16 Sets the bit depth to 16 bit floating point. Use for film sessions using OpenEXR, DPX and Kodak Cineon files. Float 16 will have slightly less precision than Float 32 but will take up much less memory

32 bit float : GPU hardware render
Float 32 Sets the bit depth to 32 bit floating point. Use for film sessions using OpenEXR, DPX and Kodak Cineon files. Float 32 will be higher precision than Float 16, but will take up much more memory

Disabled : CPU software render
The GPU is bypassed with this setting. Useful for headless render systems that do not include a GPU

2. What is the difference between Compression and Camera Noise. I’ve left it at Compression+Noise so far.

The Compression model was trained to work on reducing compression artifacts
The Camera Noise model was trained to reduce image sensor noise from low light / fast ASA settings

3. Hardware usage during rendering in Vegas Pro (1080 25P 8bit) using AMD VCE, there is almost no GPU usage outside of screen display with usage between 3-7%, and CPU usage between 5 and 49%. This results in a 3fps rate of rendering. Curious as to whether there will be further hardware refinements and improvements in the future.

We have a focused engineering team that is constantly working on refining not only the image result quality but also the processing performance on different GPU’s and operating systems.

@Former user are you referring to DeNoise ML? I've carefully compared when using it with the original 1080 noisy footage to see if there is any softening on faces, hands, etc., and have not noticed any myself.

I can't answer for 480P footage.

@Reyfox

It appears to have a frame by frame dynamic decompression algorithm. I was doing a comparison with Neat Video NR, zoomed up on a face, for 1 frame, Neat video did not touch the eye brow or the eye, it got compression artifacts out of eyelid and eye crease, making it generally look better. On that frame compression+noise removed decompression noise from eye crease making it look better but it softened eye lid.

Next frame with more compression artifacts Neat Video made eye lid very slightly softer, did not cause any other problems. Compression + noise destroyed that frame, soft eye, eyebrow, eye crease, eye lid, and it looks to apply that level of softness to entire frame it's not working on a specific area.

I've found, without changing any of the stock parameters with DeNoise ML, that it works exceedingly well at removing noise in the footage I've tried it on. No, whatever clip it is applied will not play back smoothing at all on the timeline like the Vegas Denoise

There's not really any settings to change, except and a little suspiciously the blend with original footage slider. That is perhaps a way of getting some detail back when compression + noise beauty filters everything. Also I think i'm being charitable saying beauty filter, because good quality beauty filters track the face and know what to soften and what not to, compression + noise does not seem to know what a face looks like, and doesn't behave that way.

All this relates to the single 480P video in previous post. It's high contrast due to being telecined from 16mm film back in the 80's. Maybe that makes a difference too.

@Former user have you tried it with higher resolution video? The video I used I received from a friend. It was shot at a dance event. The scene was really dark, and when looking at the video, quite noisy. As for the detail in it, it wasn't well defined, but I looked at the hand movements along with the same on a 1080 monitor, full screen, Best>full setting. Going frame by frame and comparing I saw no "softening". Maybe because a lot of the features were soft already.

Either way, since I use Boris Continuum a lot, I find the addition of DeNoise ML a great addition. I haven't tried UpRes ML yet. Next on the test list. Also, with Boris saying they will continue to work on it, is a plus. Otherwise, I would have probably bought NeatVideo 4K. Updating Continuum around Black Friday and getting all the updates to the plugins is about the same cost as buying NeatVideo alone.

@Reyfox I tried some 8bit 1080P (AR 4:3) drone footage where I color graded to my preferred look but compression degraded severely in darker areas. It had ISO noise already, and now blocky compression artifacts plus noise.

In that situation it didn't really do much at all, I think better than original, but nothing like the Neat Video result, which still contained noise but it was acceptable. Original(graded) and compression+noise was no good. I bet it's good in the right conditions, just not as versatile as Neat Video. also compression+noise encoded at 3.5fps, Neat Video at 56fps NVENC via Voukoder