Deinterlacing the fastest and easiest way

[bootroom]

I deal with lots of 1080i 25fps source videos. i need smart deinterlacing method for 720p 50fps output. don't need anything fancy with qtgmc etc.

suggest me best settings for that purpose please

[johnmeyer]

tdeint()

or

yadif()

[manolito]

Totally agreed...

For double rate output (bob deinterlace) both plugins need the "mode=1" parameter, their default is same rate output. And the regular Yadif plugin by Fizick is a so called "C-Plugin". This means that it will not autoload under plain vanilla AviSynth, you have to load it manually with the "LoadCPlugin" command. Or you create a text file "Yadif.avsi" in the Avisynth\Plugins folder which contains just the following line: LoadCPlugin("yadif.dll").

Cheers
manolito

[Motenai Yoda]

(avs+ x64)
Yadif(mode=1)
1 thread - 1096 fps
2 threads - 1959 fps
4 threads - 2775 fps

Yadifmod2(mode=1)
1 thread - 1374 fps
2 threads - 2280 fps
4 threads - 3040 fps

nnedi3(-2)
internal 4 - 248 fps
internal 8 - 265 fps
mt 2 - 154 fps
mt 4 - 227 fps

yadifmod2(mode=1,edeint=nnedi3(-2))
1 thread + 8 nnedi3 internal - 248 fps
1 threads + 4 nnedi3 internal - 236 fps
2 threads + 4 nnedi3 internal - 238 fps
4 threads + 2 nnedi3 internal - 138 fps
4 threads + 4 nnedi3 internal - 203 fps

tdeint(mode=1)
4 threads - 496 fps
tdeint(mode=1,slow=0,type=0,mtnmode=0,ap=25,aptype=1)
4 threads - 845 fps
tdeint(mode=1,slow=0,type=0,mtnmode=0)
4 threads - 1150 fps
tdeint(mode=1,slow=0,type=0,mtnmode=0,edeint=nnedi3(-2))
4 threads - 209 fps

qtgmc
- draft
1 thread - 273 fps
4 threads - 690 fps
- ultrafast
1 thread - 113 fps
4 threads - 272 fps

[manolito]

Deinterlacing the fastest and easiest way

edeint=nnedi3(-2)

This may still be pretty easy, but as soon as nnedi is used it will be rather slow... :devil:

[tormento]

(avs+ x64)
Curiosity: can you give a estimation of output quality too? Same rate (not double fps) and denoise not needed.

[manolito]

@ Motenai Yoda

I saw that you edited your post adding detailed benchmarks, thanks.

But your benchmarks confirm what I said before:
Yadif and YadifMod2 and maybe also TDeint can be called "fast". Everything else runs with less than one tenth of the speed of Yadif, and in my view this this awfully slow.

But this is for the OP to decide what he considers "fast" or not...


Cheers
manolito

[poisondeathray]

In real world , you probably will be limited by decoding speed. Your actual 1080i25 source probably won't decode at 2000-3000+ fps . But those benchmarks are a good estimate of which deinterlacer is "faster"

And if you use one of the "fast" deinterlacers , your bottleneck will probably be the resizer when going to 1080=>720, not the deinterlacer

[wonkey_monkey]

In real world , you probably will be limited by decoding speed. Your actual 1080i25 source probably won't decode at 2000-3000+ fps . But those benchmarks are a good estimate of which deinterlacer is "faster"

Does that not raise the question of what source was used, then? If it wasn't a "real world" example it could give misleading benchmarks.

[Motenai Yoda]

a ntsc dvd was used

Curiosity: can you give a estimation of output quality too? Same rate (not double fps) and denoise not needed.
deinterlace quality is pretty much source dependent

[bootroom]

Source is basically tennis matches taken from satelite.

Does leaving yadif(),tdeint() empty in bracket means it will add automatic settings for the files?

[StainlessS]

Does leaving yadif(),tdeint() empty in bracket means it will add automatic settings for the files?

Yes, uses the default settings.

[manolito]

Yes, uses the default settings.

Which means that the 25fps interlaced source will become a 25fps progressive clip. If the OP wants a 50fps progressive result (which he indicated in his first post) then "mode=1" is required for both deinterlacers.


Cheers
manolito

[bootroom]

Which means that the 25fps interlaced source will become a 25fps progressive clip. If the OP wants a 50fps progressive result (which he indicated in his first post) then "mode=1" is required for both deinterlacers.


Cheers
manolito


yeah i want 50p output. so just the mode=1 for it?

so
DirectshowSource("filename")
yadif(mode=1)

any other value to make it better but not slow it down much?

[StainlessS]

Why dont you just try variations with your clip,
if certain arg settings always made things better and faster, there would be little point in them being user configurable.
[As nobody else has your clip, you are in the best position to do that].

[Selur]

Looking at gpu based decoders which can deinterlace might also be worth a try:
In case you have a NVIDIA graphic card using DGDecNV and it's deinterlacer might be a good choice, assuming the source can be decoded through DGDecNV.
FRIM might also be worth a look.

[videoh]

assuming the source can be decoded through DGDecNV Actually, that's not necessary. The utility filters are stand-alone. You can invoke nVidia deinterlacing with PVBob(), or the CUDA version of Yadif with DGBob(). They can be used with any source filter. You do need to load DGDecodeNV.dll, of course, but you don't have to use DGSource().

[hydra3333]

Assuming no HDR ? ;)

[videoh]

Not sure what point you are making. DGSource() supports HDR and so do several other source filters.

[Selur]

@hydra3333: Why would you assume bootrooms "1080i 25fps source videos" are HDR?

[hydra3333]

Not sure what point you are making. DGSource() supports HDR and so do several other source filters.
OK. I wasn't sure all of the DG filters did HDR just yet. If they do then that is marvellous !! Seriously.

@hydra3333: Why would you assume bootrooms "1080i 25fps source videos" are HDR?
OK, I didn't assume because I'd been caught short recently with something that was purported to be "HDR" and actually wasn't. HDR wasn't explicitly stated that I'd noticed (happy to be shown to be wrong about that) and HDR is seems to be gradually seeping into things nowadays, so it seemed best not to make an assumption it wasn't HDR.

[videoh]

OK. I wasn't sure all of the DG filters did HDR just yet. If they do then that is marvellous !! Seriously.
Currently, DGSource(), DGDenoise(), and DGSharpen() support high bit depth.