Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
4k denoise: neatvideo versus in camera noise reduction
  • Neatvideo render time in 4k video is very long, very slow render, very time consuming, but quality is better than in camera noise reduction, so...

    Two options for 4k denoise:

    1 - use noise reduction "-5" in camera and use neatvideo in post production for a better quality denoise.

    2 - use noise reduction "0" in camera and avoid neatvideo in post-production for a faster workflow.

    Which one to choose?

  • 10 Replies sorted by
  • No in-camera NR is doing temporal reduction. Neatvideo is slow, yes, but will preserve image detail much better.

    So, either give your camera more light or use NV on critical shots.

  • Decisions are more related to time.

    My quad-core computer with average nvidia card is taking 24x the footage time to do neatvideo denoise in 4k.

    An octa-core at 4.5ghz with nvidia 1080 will take more or less 12x the footage time to do neatvideo denoise in 4k.

    So if shooting weddings, parties, interviews, long narrative episodes, things with near deadline, the best thing is to use in-camera noise reduction.

    If shooting small footage or long things without near deadline, so neatvideo can be a better quality option.

  • @apefos It should be a bit less than that with an octa-core 4.5ghz and a 1080. My quad-core 4ghz and gtx 1080 are closer to 1/7 or 1/8 realtime (and it might be a bit less if I closed Chrome in the background).

    8 cores in the benchmark is actually referring to the 8 threads on a quad-core i7. Assuming that performance scales in a linear fashion, I'd expect another 4 slightly faster cores to add about 1.2 frames per second. A 1080ti might get you another 1 as well - so you'd be down to around 6x the time (but that's still a lot slower, of course).

    Have you tried the interframe noise reduction in Resolve? If theirs is faster, you could just apply it to your footage before ingesting it into whatever other editor you're using.

    neat_profiling.png
    802 x 582 - 66K
  • Maybe the temporal (interframe) in neatvideo is slowing the process, the G7 needs temporal noise reduction with some spatial.

  • I'm not sure where the attachment to my comment went before - I re-added it so that the second paragraph would make more sense. That's the benchmark for Neat Video on my system (4x4/gtx 1080) - about 3.2fps.

  • Maybe somebody with a full license for resolve could say what sort of performance its denoiser gives... That or maybe Red Giant's is faster? I think I have a copy of it somewhere. I can give it a shot.

  • The latest versions of Neat Video use the CPU and the GPU, either separately or in combination. Run the optimizer to find out your frame rate. If you don't have at least a hexacore CPU and a Graphics card with a lot of CUDA cores, these two simple upgrades will make a big difference.
    I sleep at night a few hours, so if I need a monster project rendered, I just queue it up the jobs to render at night.
    It's worth noting that running Neat video on 4K is going to be a lot slower than 1080p--so many more pixels.

  • I just asked a buddy - he didn't have exact numbers handy, but he said that Resolve with interframe gets about 10fps on his 1080ti/8x4.8ghz monster system... so it's at least a bit faster.

    I'm doing a test now with Red Giant Denoiser II. I have it on an older computer (2015 iMac 5k (quad-4ghz, R9 395X)) and a 28:27 second clip is showing about 4:30 left at the halfway point (about 1/18 speed) where a similar export shows about 28 seconds left at the halfway point (about 1/2 speed) - so RG is about as slow as NV with similar settings... though RG also has more options to reduce denoiser quality, etc, which might let you speed things up a bit.

  • The neatvideo benchmark shows faster results compared with render times.

    my benchmark best combination for 4k radius 4 is 1,36fps, but render is 1fps (24x the 2160p24 footage time)

    my benchmark best combination for 4k radius 1 is 2,23fps, but render is 1,5fps (16x the 2160p24 footage time)

    For my eyes NeatVideo quality is awesome, so if render time is not a problem it worth the effort.

  • Today I did careful tests comparing neatvideo versus in-camera noise reduction.

    I did shots in exterior shadow+sun in same scene and also interior low light light+shadow.

    The tests was done in iso 1600 with shadow+2 and highlight-2 with the G7 in 4k.

    In-camera noise reduction was tested from -5 to +5 in each scene, a total of 22 takes.

    Noise reduction -5 from the two scenes was denoised with neatvideo with the presets and profiles developed with the noise target.

    Conclusions:

    The best in-camera noise reduction in both scenes was the +1 because it is the maximum amount of noise reduction before the image starts to get soft.

    In exterior scene it worked better, removing almost all luma noise in the shadows.

    In interior scene the in-camera noise reduction was not useful at all, because the amount of fine noise was impossible to remove even in +5. It was minimized but not removed.

    In both exterior and interior scenes the image with in-camera noise reduction keeps trembling, the image dance after noise minimized, and chroma noise was not removed and keeps floating in the screen.

    So for a professional result and a beautiful 4k image the in camera noise reduction is bad choice.

    NeatVideo is just excellent. Removes both luminance and chrominance noise without residual, the image does not tremble after denoise. The 1600 iso denoise with neatvideo looks better than iso 200 without denoise. The image resolution keeps perfect without any soft, with pleasing texture, without plastic look. The results is just professional, can compete with high end cameras.