Santiag for progressive to interlaced?

[marcorocchini]

Hi

please be patient:
# <--here a poor cat ###

So my target is transcode properly my horrible drone that is a 25fps, output format: MXF xdcamhd422 profile, 50i interlaced

whit a filter to avoid aliasing ad artifacts due to conversion.

Please download the horrible:

https://www.swisstransfer.com/d/45e6c1bf-f64c-4af1-a73b-c96718efe466

My empirically script (drone7.avs):

Import("v:\automazioneclip\avisynth\plugins\IResize.avsi")
Import("V:\automazioneclip\AviSynth\plugins\smoothFPS2.avsi")
Import("V:\automazioneclip\AviSynth\plugins\smoothFPS3.avsi")
Import("v:\automazioneclip\AviSynth\plugins\CropResizedic2017.avsi")
Import("V:\automazioneclip\AviSynth\FrameRateConverter2021\FrameRateConverter.avsi")
LoadPlugin("v:\automazioneclip\AviSynth\FrameRateConverter2021\X64\FrameRateConverter.dll")
LoadPlugin("v:\automazioneclip\AviSynth\colormatrix26\X64\ColorMatrix64.dll")
Import("V:\automazioneclip\AviSynth\FFMS2_AVSI_2019\ffms2.avsi")
LoadPlugin("v:\automazioneclip\AviSynth\LSMASH_AGO_2023\x64\LSMASHSource.dll")
LoadPlugin("V:\automazioneclip\AviSynth\FFMS2_R1363_AGO2023\X64\ffms2.dll")
LoadCPlugin("v:\automazioneclip\avisynth\plugins64\yadif.dll")
LoadPlugin("V:\automazioneclip\AviSynth\MaskTool2\X64\masktools2.dll")
LoadPlugin("V:\automazioneclip\AviSynth\RIFE\avsresize.dll")
LoadPlugin("v:\automazioneclip\AviSynth\BestSource\BestSource.dll")
LoadPlugin("v:\ResampleMT.dll")
LoadPlugin("v:\nnedi3.dll")
LoadPlugin("v:\aWarpsharpMT.dll")
#LoadPlugin("v:\nnedi2.dll")
LoadPlugin("v:\eedi3.dll")
LoadPlugin("v:\EEDI2.dll")
LoadPlugin("v:\GScript_26_64.dll")
LoadPlugin("v:\SangNom2.dll")

Import("c:\santiag.avsi")

BSVideoSource("v:\25fpsDrone.mp4")
#ConvertToYV24()
FrameRateConverter(50000,1000, preset = "faster")

santiag(strh=2, strv=1, type="nnedi3", nns=0, threads=0, nsize=6, halfres=false, scaler_post="Spline36Resize")

ConvertToYUY2(interlaced=false)

assumetff().separatefields().SelectEvery(4, 0, 3).weave() #interlace

Prefetch(8)

and then the transcoding process is:

ffmpeg2025.exe -v quiet -stats -y -i v:\drone7.avs -an -pix_fmt yuv422p -f rawvideo - | v:\ffmbcX642017.exe -threads 12 -s 1920x1080 -r 25 -y -f rawvideo -pix_fmt yuv422p -i - -i "v:\10 - Traccia 10_HDavs.wav" -tff -target xdcamhd422 -an -ac 4 v:\drone5.mxf -newaudio -acodec pcm_s24le -ar 48000

processing is slow 7 fps

this my results:
https://www.swisstransfer.com/d/c69054c8-c470-48c5-9d9e-5d8823c6fcbc

I try to replace emask with santiag to see if I can see better results, or I'm wrong? in any case can the script be better corrected?
let me know your opinion cat

thanks thanks

[Sharc]

Use RIFE

LWLibavVideoSource("25fpsDrone.mp4") #your progressive 25fps source

RGB=z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:auto:auto:limited=>rgb:same:same:full") # convert to RGB floats
RIFE=RGB.RIFE(gpu_thread=1, model=11, fps_num=50000, fps_den=1000, sc=true, sc_threshold=0.3) # motion interpolation using model 11
YV12=RIFE.z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:auto:auto:full=>709:same:same:limited") # convert back to YV12

interlace=YV12.separatefields().selectevery(4,0,3).weave() #interlace

prefetch(8)
return interlace

[Sharc]

^^^^
Improvement: Select model=12 in the script above. It seems to produce less annoying interpolation artefacts, but is slower. You can experiment yourself to find the "best" model ....

LWLibavVideoSource("25fpsDrone.mp4")  #your 25p source

RGB=z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:auto:auto:limited=>rgb:same:same:full") # convert to RGB floats
RIFE=RGB.RIFE(gpu_thread=1, model=12, fps_num=50000, fps_den=1000, sc=true, sc_threshold=0.3) # motion interpolation using model 12
YV12=RIFE.z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:auto:auto:full=>709:same:same:limited") # convert back to YV12

interlace=YV12.assumeTFF().separatefields().selectevery(4,0,3).weave() #interlace

prefetch(8)
return interlace

[marcorocchini]

Sharc cat for kindness:

my target is encode to 422 (xdcamhd422) so I try this:

BSVideoSource("v:\25fpsDrone.mp4")  #your 25p source

RGB=z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:auto:auto:limited=>rgb:same:same:full") # convert to RGB floats
RIFE=RGB.RIFE(gpu_thread=1, model=12, fps_num=50000, fps_den=1000, sc=true, sc_threshold=0.3) # motion interpolation using model 12
YV12=RIFE.z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:auto:auto:full=>709:same:same:limited").convertToYUY2(interlaced=false) # convert back to YV12


interlace=YV12.assumeTFF().separatefields().selectevery(4,0,3).weave() #interlace


return interlace
prefetch(8)

need I to change the YUV420P8 parameter?

and is there a more speed model?

[Sharc]

For YUV422 output:

LWLibavVideoSource("Drone\25fpsDrone.mp4")

RGB=z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:auto:auto:limited=>rgb:same:same:full") 
RIFE=RGB.RIFE(gpu_thread=1, model=12, fps_num=50000, fps_den=1000, sc=true, sc_threshold=0.3) 
YV16=RIFE.z_ConvertFormat(pixel_type="YUV422P8", colorspace_op="rgb:auto:auto:full=>709:same:same:limited")
interlace=YV16.assumeTFF().separatefields().selectevery(4,0,3).weave()

prefetch(8)
return interlace

...and is there a more speed model?

Try all models, one by one, and tell us your findings.

[marcorocchini]

model 12 seems good, but maybe, I need to apply a little, advanced, antialiasing line.

However:

let explain the cat

This is my source:

https://www.swisstransfer.com/d/5eaa2e33-8ac3-4f34-aea4-f14062ca14f1

it's my low quality drone video wow, 25P

My target is get an interlaced version of it, 50i, transcoded ok, minimizing artifacts.

I tryied with Da*****R*****ve: import a 50i in timeline-->import the horrible drone 25P file--> set the use of Optical Flow --> set the SuperScale 4X processing. From this I get this output:

https://www.swisstransfer.com/d/f845b80a-3039-469a-beb5-2093d517f8c3

that seems not bad but there are some little artifacts in certain point.

However I don't like a lot to use Da*****R*****ve, if possible, and I try to transcode with an avisynth script.

So what is the correct way to optimize this process?

[Sharc]

I am not aware of any interpolator which does NOT produce any artifacts.
RIFE is sharper, preserves much more details and has much less interpolation artifacts (see for example the TV antenna on the roof) than your Resolve version.
Watch the separated fields of the 2 variants interleaved and you will easily see the differences.

[marcorocchini]

I cat understand

but the result of the only rife seems a little too sharp for me, is there perhaps some rife models that integrates something like greater antialiasing?

The preservation of details, however important, risks becoming excessive if this implies a little too much "noise" on images that have a low quality origin like my drone, or images coming from low level cameras whose electronic signal processing is not at low intermodulation, in this case, at this point, it seems more correct to "hide" details during transcoding even if overall the image is slightly softer, rather than maintaining high details but at the same time high crudeness and crispening

[Sharc]

You can always soften or blur or otherwise tweak your video (e.g. just before re-interlacing the RIFE interpolation) using usual Avisynth filters for example. Or you can apply filters to your original progressive source. Try what suits you best.

[marcorocchini]

ok cat but there are a recent antialiasing plugin for avisynth? or something that apply a little noise reduction?

What can it be the final script for my horrible drone source?

[Sharc]

What can it be the final script for my horrible drone source?

The one which pleases your eyes most.

If you want to reduce "crispiness" and aliasing you can try on progressive video something like

LWLibavVideoSource("25fpsDrone.mp4")

RGB=z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:auto:auto:limited=>rgb:same:same:full")
RIFE=RGB.RIFE(gpu_thread=1, model=12, fps_num=50000, fps_den=1000, sc=true, sc_threshold=0.3)
YV12=RIFE.z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:auto:auto:full=>709:same:same:limited") # convert back to YV12

mask=masktools2_mt_edge(YV12).grayscale()
blur=YV12.blur(1.0)
res=overlay(YV12, blur,mask=mask,opacity=1.0,mode="blend")

interlace=res.assumeTFF().separatefields().selectevery(4,0,3).weave() #interlace

prefetch(8)

return interlace

[marcorocchini]

sharc cat please I need that: the script should present an output YUV422

Need I to add at the end of the script the converttoYUY2(interlaced=false) ?

[Sharc]

sharc cat please I need that: the script should present an output YUV422

Need I to add at the end of the script the converttoYUY2(interlaced=false) ?

No.
I already told you in post#5 how to do it. So here again:

LWLibavVideoSource("25fpsDrone.mp4")

RGB=z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:auto:auto:limited=>rgb:same:same:full")
RIFE=RGB.RIFE(gpu_thread=1, model=12, fps_num=50000, fps_den=1000, sc=true, sc_threshold=0.3)
YV16=RIFE.z_ConvertFormat(pixel_type="YUV422P8", colorspace_op="rgb:auto:auto:full=>709:same:same:limited")

YV16=YV16.santiag(3,3)
mask=masktools2_mt_edge(YV16).grayscale().blur(0.5)
blur=YV16.blur(1.0)
res=overlay(YV16, blur,mask=mask,opacity=1.0,mode="blend")

interlace=res.assumeTFF().separatefields().selectevery(4,0,3).weave()

prefetch(8)
return interlace

[marcorocchini]

Sharc I think to have a problem with all version of Avisynth greater than R3661: I use Mainconcept Totalcode to encode the .avs but the versions that support RIFE wont load the audio part of the .avs

I don't know the reason but I force to use, at leat for the moment, Avisynth R3661 and RIFE is not working.

So I try to replace RIFE with the smoothFPS2.avsi:

function SmoothFPS2(clip source, int num, int den) {
super = MSuper(source, pel=2, hpad=0, vpad=0, rfilter=4)
backward_1 = MAnalyse(super, chroma=false, isb=true, blksize=32, searchparam=3, plevel=0, search=3, badrange=(-24))
forward_1 = MAnalyse(super, chroma=false, isb=false, blksize=32, searchparam=3, plevel=0, search=3, badrange=(-24))
backward_2 = MRecalculate(super, chroma=false, backward_1, blksize=8, searchparam=1, search=3)
forward_2 = MRecalculate(super, chroma=false, forward_1, blksize=8, searchparam=1, search=3)
backward_3 = MRecalculate(super, chroma=false, backward_2, blksize=4, searchparam=0, search=3)
forward_3 = MRecalculate(super, chroma=false, forward_2, blksize=4, searchparam=0, search=3)
MBlockFps(source, super, backward_3, forward_3, num, den, mode=0)}

that seems work not very bad.

but I'm not able to modify the script to use it instead of Rife, what can I do?

[Sharc]

Your posted source file '25fpsDrone.mp4' has no audio track, btw.

But anyway, here the RIFE replacement as you wish:

LWLibavVideoSource("25fpsDrone.mp4")

YV16=z_ConvertFormat(pixel_type="YUV422P8", colorspace_op="709:auto:auto:limited=>709:same:same:limited")
YV16=SmoothFPS2(YV16,50000,1000)

YV16=YV16.santiag(3,3)
mask=masktools2_mt_edge(YV16).grayscale().blur(0.5)
blur=YV16.blur(1.0)
res=overlay(YV16, blur,mask=mask,opacity=1.0,mode="blend")
interlace=res.assumeTFF().separatefields().selectevery(4,0,3).weave() #interlace

prefetch(8)
return interlace

################# FUNCTION ####################################################
function SmoothFPS2(clip source, int num, int den) {
super = MSuper(source, pel=2, hpad=0, vpad=0, rfilter=4)
backward_1 = MAnalyse(super, chroma=false, isb=true, blksize=32, searchparam=3, plevel=0, search=3, badrange=(-24))
forward_1 = MAnalyse(super, chroma=false, isb=false, blksize=32, searchparam=3, plevel=0, search=3, badrange=(-24))
backward_2 = MRecalculate(super, chroma=false, backward_1, blksize=8, searchparam=1, search=3)
forward_2 = MRecalculate(super, chroma=false, forward_1, blksize=8, searchparam=1, search=3)
backward_3 = MRecalculate(super, chroma=false, backward_2, blksize=4, searchparam=0, search=3)
forward_3 = MRecalculate(super, chroma=false, forward_2, blksize=4, searchparam=0, search=3)
MBlockFps(source, super, backward_3, forward_3, num, den, mode=0)}
###############################################################################

[marcorocchini]

sharc cat for kindness: seems that the script output is YV16.

Can I get it YUY422 for direct encoding to ffmbc/ffmpeg/totalcode?

[Sharc]

YV16 is YUV 4:2:2 planar. Your YV12 4:2:0 source is converted to YUV 4:2:2 planar in line 2 of the script of post#15.
If you need it interleaved instead of planar you can replace the last line of the script by

return interlace.ConvertToYUY2(interlaced=true)

This conversion is lossless.

[marcorocchini]

sharc cat I ask this:

due to the fact I can't use, at the moment, Rife

Does exist an alternative to SmoothFPS2 that do sometimes artifacts?

[Sharc]

sharc cat I ask this:

due to the fact I can't use, at the moment, Rife

Does exist an alternative to SmoothFPS2 that do sometimes artifacts?

I don't understand your question, but RIFE is better than SmoothFPS2 and better than the other mvtools based interpolators I know (like framerateconverter) and better than what you produced with Resolve. "Better" means that it produces less annoying artefacts.
You may google for other alternatives.

[jagabo]

There are several motion compensated frame rate interpolators like FrameRateConverter(), InterFrame(), SVPflow(), etc.

http://www.avisynth.nl/index.php/External_filters#Frame_Rate_Conversion

Most variations use mvtools.

[marcorocchini]

Please help the poor cat:


[Attachment 86670 - Click to enlarge]

I would like to "incapsulate" in 1 unique function the 3 lines:

santiag(2,2)
fmask=masktools2_mt_edge().grayscale().blur(0.5)
overlay(last, blur(1,0),mask=fmask,opacity=1.0,mode="blend")

Do I need compile an .avsi file?

function SantiagComplex(clip c)

{

santiag(2,2)
fmask=masktools2_mt_edge().grayscale().blur(0.5)
overlay(last, blur(1,0),mask=fmask,opacity=1.0,mode="blend")

}

But seems it dont work

[Sharc]

What's the purpose?

Didn't test it much, so try below

function SantiagComplex(clip c,int radius,float decrisp)
{
santiag(c,radius,radius)
mask=masktools2_mt_edge(c).grayscale().blur(0.5)
overlay(c, blur(c,decrisp),mask=mask,opacity=1.0,mode="blend")
}

Call it in the script post#15 (line res=....) like

res=SantiagComplex(YV16,2,1.0)

And eventually try:

LWLibavVideoSource("D:\Temp\marcorochini - the cat\Drone\25fpsDrone.mp4")

z_ConvertFormat(pixel_type="YUV422P8", colorspace_op="709:auto:auto:limited=>709:same:same:limited")
SmoothFPS2(last,50000,1000)
SantiagComplex(last,2,1.0)

assumeTFF().separatefields().selectevery(4,0,3).weave() #interlace
ConvertToYUY2(interlaced=true)  #for 422 interleaved output

prefetch(8)


########### FUNCTIONS #########################################################################################
function SmoothFPS2(clip source, int num, int den) {
super = MSuper(source, pel=2, hpad=0, vpad=0, rfilter=4)
backward_1 = MAnalyse(super, chroma=false, isb=true, blksize=32, searchparam=3, plevel=0, search=3, badrange=(-24))
forward_1 = MAnalyse(super, chroma=false, isb=false, blksize=32, searchparam=3, plevel=0, search=3, badrange=(-24))
backward_2 = MRecalculate(super, chroma=false, backward_1, blksize=8, searchparam=1, search=3)
forward_2 = MRecalculate(super, chroma=false, forward_1, blksize=8, searchparam=1, search=3)
backward_3 = MRecalculate(super, chroma=false, backward_2, blksize=4, searchparam=0, search=3)
forward_3 = MRecalculate(super, chroma=false, forward_2, blksize=4, searchparam=0, search=3)
MBlockFps(source, super, backward_3, forward_3, num, den, mode=0)}

function SantiagComplex(clip c,int radius,float decrisp)
{
santiag(c,radius,radius)
mask=masktools2_mt_edge(c).grayscale().blur(0.5)
return overlay(c, blur(c,decrisp),mask=mask,opacity=1.0,mode="blend")
}
############################################################################################################

[marcorocchini]

ah wow seems works

SantiagComplex(YV16,2,1.0)

but I wonder:

the 1.0 is the value for blur argument?

In this case, the equivalent of the original line

santiag(2,2)

what should be using the new SantiagComplex function?

[Sharc]

... the 1.0 is the value for blur argument?

Yes. This is the blur value for making the video look less sharp and less crispy but still keeping details - as you wanted in post#8. It has nothing to do with santiag.

The script in post#22 does the same as the script in post#15. Only difference is the santiag parameters, and the "incapsulation" of santiag and the edge masked blurring in a new function which you named SantiagComplex() in post#21.
I suggested santiag(3,3) in my earlier posts, you wanted santiag(2,2) in your non-working attempt in post#21. So I changed it to (2,2) in post#22.
The function SantiagComplex(clip c,int radius,float decrisp) as defined at the end of the script in post#22, when called, invokes santiag(radius,radius) plus the edge masked blur(decrisp).
So calling like SantiagComplex(v,2,1.0) means to process the video v with santiag(2,2) and with the edge masked blur(1.0)

Just copy and paste the full script of post#15 or #22 and be done with it. They do the same, with the exception of the santiag parameters which I changed upon your request.

(And once you have RIFE working revert to post#5 or #11).

[marcorocchini]

ah wow


[Attachment 86718 - Click to enlarge]