Need advice how to fix using Avisynth this prog video with interlaced artif

[elektro]

QTGMC alone can't do a proper job with messed up fields. My main concern is to make the scanlines disappear as much as possible.

[jagabo]

A smooth horizontal panning should would help analyze the problems. But the first thing you need to do is remove the duplicate frames with TDecimate().

[elektro]

There are some duplicate interlaced frames which cause backward jumps in motion. TDecimate() gets rid of those first:

LSMASHVideoSource("Aurora - Lollapalooza Chile, Chile, 2023.mp4")
AssumeTFF()
TDecimate(Cycle=5, CycleR=1)
Blur(0.0, 0.25)
QTGMC()

A slight vertical blur (while interlaced) cleans up a lot of the inter-field artifacts. Followed by A QTGMC() to deinterlace. There are still missing frames (which can be see as jumps in motion). Some smart-ish motion interpolation should help with that. A smooth panning show whould would help with that.

Thanks. Additionally I resized down the video frame by 50% to make the artifacts less noticeable.

[jagabo]

That script worked for some frames but didn't seem to work for others. That's why I removed it from the earlier post. A sharp, smooth, horizontally panning shot is needed for better analysis.

[elektro]

That script worked for some frames but didn't seem to work for others. That's why I removed it from the earlier post. A sharp, smooth, horizontally panning shot is needed for better analysis.

I uploaded another clip. That's the best I could find. I'll use those Avisynth commands for the entire recording, so it's important to get them right (to get the best result based on the source video).

[jagabo]

I didn't really come up with anything that cleaned up the residual artifacts without damaging the picture too much. But this fixes the motion for the clip in post #5:

###################################################

function SmoothFPS_RIFE(clip source, int num, int den)
{ 
    source
    z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:709:709:l=>rgb:709:709:f")
    Rife(gpu_thread=1, model=9, fps_num=num, fps_den=den, sc=true, sc_threshold=0.12)
    z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:709:709:f=>709:709:709:l")
}

##################################################

LSMASHVideoSource("Aurora - Lollapalooza Chile, Chile, 2023-2.mp4") 
AssumeTFF()
TDecimate()
QTGMC()
# a strong temporal filter here may remove move some residual noise
SmoothFPS_RIFE(48000*3, 1001)
SelectEvery(24, 0, 3, 4, 5, 6, 9, 12, 15, 18, 21)

RIFE is very slow so you might use an alternate frame rate converter, like FrameRateConverter() or InterFrame(). They don't work as well though.

You may have to change the SelectEvery() arguments as parts of the video have a different "phase" of missing frames.

[elektro]

I didn't really come up with anything that cleaned up the residual artifacts without damaging the picture too much. But this fixes the motion for the clip in post #5:

###################################################

function SmoothFPS_RIFE(clip source, int num, int den)
{ 
    source
    z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:709:709:l=>rgb:709:709:f")
    Rife(gpu_thread=1, model=9, fps_num=num, fps_den=den, sc=true, sc_threshold=0.12)
    z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:709:709:f=>709:709:709:l")
}

##################################################

LSMASHVideoSource("Aurora - Lollapalooza Chile, Chile, 2023-2.mp4") 
AssumeTFF()
TDecimate()
QTGMC()
# a strong temporal filter here may remove move some residual noise
SmoothFPS_RIFE(48000*3, 1001)
SelectEvery(24, 0, 3, 4, 5, 6, 9, 12, 15, 18, 21)

RIFE is very slow so you might use an alternate frame rate converter, like FrameRateConverter() or InterFrame(). They don't work as well though.

You may have to change the SelectEvery() arguments as parts of the video have a different "phase" of missing frames.

Thanks. I appreciate your efforts to help. My PC is pretty old and it will take ages to process the entire video. I'd rather prefer to keep the actual frames (no double frame rate, no interpolated frames), just to remove the interlacing artifacts as much as possible even at the cost of reduced picture quality (but not by much). Is this the optimal solution or is there a better way?

LSMASHVideoSource("Aurora - Lollapalooza Chile, Chile, 2023.mp4")
AssumeTFF()
TDecimate(Cycle=5, CycleR=1)
Blur(0.0, 0.25)
QTGMC()

[davexnet]

What is the source of this video? DO you have the camera files?
It's a shame, nice looking video but mangled by somebody who didn't know any better.
We see that too often on this site

[elektro]

What is the source of this video? DO you have the camera files?
It's a shame, nice looking video but mangled by somebody who didn't know any better.
We see that too often on this site

I have the entire video (1.4GB). It seems like it was recorded from a livestream. That bloody interlacing saved as progressive video ruined the quality, especially in combination with a fairly low bitrate for the AVC format. I'm trying to get rid of the interlacing artifacts. I also have a Brazil video with a much better quality. I collect concert videos and music videos.

[jagabo]

I worked something that cleans up a lot of the residual noise without damaging the picture too badly:

function SmoothFPS_RIFE(clip source, int num, int den)
{ 
    source
    z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:709:709:l=>rgb:709:709:f")
    Rife(gpu_thread=1, model=9, fps_num=num, fps_den=den, sc=true, sc_threshold=0.12)
    z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:709:709:f=>709:709:709:l")
}

###################################################
#
# build a mask of areas where there are 
# alternating horizontal lines
#
##################################################

function CombMask(clip v, int "threshold")
{
    threshold = default(threshold, 5)

    Subtract(v, v.blur(0.0, 1.0).Sharpen(0.0, 0.6))
    GeneralConvolution(0, "
        0  8  8  8  0
       -0 -8 -8 -8 -0 
        0  8  8  8  0
       -0 -8 -8 -8 -0
        0  8  8  8  0", chroma=false, alpha=false)
    mt_lut("x 125 - abs")
    mt_binarize(threshold) # threshold
    mt_inpand()
    mt_expand()
    mt_expand(chroma="-128")
}

##################################################

LSMASHVideoSource("Aurora - Lollapalooza Chile, Chile, 2023-2.mp4") 
AssumeTFF()
TDecimate()
cmask = CombMask(10).BinomialBlur(3).SelectEvery(1,0,0)
QTGMC()
Overlay(last, QTGMC(InputType=2, EZDenoise=10).Blur(0.0, 1.5).aWarpSharp2(depth=10).Sharpen(0.6), mask=cmask)
SmoothFPS_RIFE(48000*3, 1001)
SelectEvery(24, 0, 3, 4, 5, 6, 9, 12, 15, 18, 21)

Here I build a mask of areas where there is combing in the original video. Then I use that mask to overlay a heavily filtered version of the video over just those areas. Normally I would never use QTGMC's EZDenoise with such high settings. But it works well here since we are limiting it to the areas with bad combing artifacts. You can play around with different EZDenoise strengths, and the additional Blur, aWarpSharp, and Sharpen settings.

[davexnet]

I asked a question about this myself in 2020.
https://forum.videohelp.com/threads/398436-In-VirtualDub-where-is-the-capture-BITRATE-...on#post2592253

Your footage seems to have other issues. Was it resized while interlaced?

[jagabo]

Yes, his video has interlaced 4:2:0 chroma and was compressed progressive. That has blended the two fields' chroma together. You can see it in panning shots and other motion.

The video show no signs of having been resized vertically. The artifacts after QTGMC are from the progressive encoding of interlaced content, the too low bitrate, and the blended chroma.

Then there's the problem of the original 30i video being decimated to 24 fps then frame repeated to 30 fps.

I don't know if any of this was caused by the OP preparing the sample clip. If so, he should provide a non-reencoded trim from the original he has.

[davexnet]

Thanks for the info jagabo. Your result with the mask overlay is impressive

[elektro]

Yes, his video has interlaced 4:2:0 chroma and was compressed progressive. That has blended the two fields' chroma together. You can see it in panning shots and other motion.

The video show no signs of having been resized vertically. The artifacts after QTGMC are from the progressive encoding of interlaced content, the too low bitrate, and the blended chroma.

Then there's the problem of the original 30i video being decimated to 24 fps then frame repeated to 30 fps.

I don't know if any of this was caused by the OP preparing the sample clip. If so, he should provide a non-reencoded trim from the original he has.

I posted the entire original clip here (click the first link): https://forum.videohelp.com/threads/412183-Need-advice-how-to-fix-using-Avisynth-this-...if#post2711030
When I posted samples, I cropped them losslessly with Avidemux.

[jagabo]

When I posted samples, I cropped them losslessly with Avidemux.

They should be fine then.

[jagabo]

Thanks for the info jagabo.

You're welcome.

Your result with the mask overlay is impressive

Thanks. Using QTGMC like that to denoise is probably not the best choice. Some other spacie-temporal NR might work better. But it ends up looking ok and you don't really notice the residual noise watching the video at 60 fps.