Does this look any good, as intra-frame noise reduction goes?

[wonkey_monkey]

Just wondered what people's thoughts might be on this example of noise reduction, as it's not something I do a lot of myself:

(the world "Original" is burned into the video so it was also subject to denoising)

Would you say that's good, bad, or mediocre? Too "glowy"? Too much lost contrast or detail, compared to the amount of denoising?

[poisondeathray]

Originally Posted by wonkey_monkey

Would you say that's good, bad, or mediocre? Too "glowy"? Too much lost contrast or detail, compared to the amount of denoising?

Yes too "glowy" - the radius of the "blur" is too high, too much detail loss

[videoh]

Maybe try an NLMeans based algorithm? KNLMeansCL, DGDenoise, etc.

[feisty2]

bad,,,

[FranceBB]

I'm kinda known for being one that filters a lot, so take my advice with a bit of salt, but I would do something like:

super = MSuper(pel=2, sharp=1)
bv1 = MAnalyse(super, isb = true, delta = 1, overlap=4)
fv1 = MAnalyse(super, isb = false, delta = 1, overlap=4)
bv2 = MAnalyse(super, isb = true, delta = 2, overlap=4)
fv2 = MAnalyse(super, isb = false, delta = 2, overlap=4)
MDegrain2(super,bv1,fv1,bv2,fv2,thSADC=800, thSAD=800)

for temporal degrain followed by a spatial one if it's needed like DFTTest:

dfttest(sigma=64, tbsize=1, lsb_in=false, lsb=false, Y=true, U=true, V=true, opt=0, dither=0)

KNLMeansCL - as suggested by others - is also kinda good and it uses the GPU dedicated memory, so it's faster.
You can start with:

KNLMeansCL(d=1, a=2, s=4, h=4, device_type="auto")

and increase step by step 'till you find your sweet spot.
Oh and by the way, all the ones above can work in 16bit planar, so I would also suggest you to work with 16bit precision to avoid banding and then use f3kdb to deband a bit before you dither down to 8bit or whatever bit depth you wanna get as output.
We can definitely get something more out of the source, but I'm curious about what you used in your screenshot. Is it Deen() or eDeen()? It looks like it, but I would advise against it as it's known to destroy details and there are better alternatives nowadays. Anyway, let me know. If you could share a little sample it would be appreciated as we can't denoise images since they're... well... static and temporal filters wouldn't work.

Cheers,
Frank

[ChaosKing]

Too much detail loss and "glowy". If you can provide a sample I'm sure you'll get some better script/filter advices.

[wonkey_monkey]

To clarify, I'm not looking for other ways to denoise this clip, but rather comments on this particular denoising. It's a sort of side effect of another filter I'm working on but I wasn't sure if it was worth spending any time on the denoising thing. Sounds like it isn't, so that will save me time. Just in case here are a few more examples with varying parameters.

[feisty2]

vaporsynth has come to save the world

import vapoursynth as vs
core = vs.get_core()
clp = core.imwri.Read(r'C:\Users\Administrator\Desktop\N5vEWfG.png', float_output=True)
clp = core.std.Crop(clp, 0, 480,0,0)

clp = core.dfttest.DFTTest(clp, smode=0, sosize=0, tbsize=1, tosize=0, tmode=0, sbsize=41, slocation=[0,1024, 0.5,0, 1,0])

ref = core.bm3d.Basic(clp,sigma=10,th_mse=10*160+1200,hard_thr=3.2,block_size=8,block_step=1,group_size=32,bm_range=24,bm_step=1)
ref = core.bm3d.Basic(ref,sigma=8,th_mse=8*160+1200,hard_thr=3.2,block_size=4,block_step=1,group_size=32,bm_range=24,bm_step=1)

clp = core.bm3d.Final(clp,ref,sigma=5,th_mse=5*120+800,block_size=4,block_step=1,group_size=32,bm_range=24,bm_step=1)
clp = core.bm3d.Final(clp,ref,sigma=4,th_mse=4*120+800,block_size=4,block_step=1,group_size=32,bm_range=24,bm_step=1)

clp.set_output()

[feisty2]

Originally Posted by FranceBB

for temporal degrain followed by a spatial one if it's needed like DFTTest:

dfttest(sigma=64, tbsize=1, lsb_in=false, lsb=false, Y=true, U=true, V=true, opt=0, dither=0)

you wouldn't be using dfttest with such silly settings if you know anything at all about frequency domain denoisers