Best Way to De-interlace HDV

[dellsam34]

I'm trying to work on some HDV files to put on YT so they don't get screwed up by YT de-interlacing algorithm, I'm taking the long way trying to preserve the quality, I want to know if there are quicker methods that result in less quality degradation, Here is what I'm doing so far:

- Input the 4:2:0 1080i HDV file into vdub2 to strip the mpeg-2 codec saving as 4:2:2 HuffYUV AVI (not sure if 10bit can make a difference here, so I'm just using 8bit).
- Input the resulting large file into AVSpmod and de-interlace with QTGMC in fast mode, followed by a resize in the second line from 1440x1080 to 1920x1080 to preserve the 16:9 aspect ratio.
- Upload the second resulting file into YT and let it do the encoding.
- Delete all the files but the original HDV source.

[hello_hello]

What's the final output format?
You should be able to open the original HDV file with an Avisynth script in VirtualDub2 or AvsPmod and re-encode and de-interlace/resize in one step, without the need for an an intermediate lossless file.

[pchan]

Just sharing what I have done to de-interlace a video clip taken from a Sony DSC-HX20V camera in 1440x1080i (16:9) 25fps AVC(h.264) to 1440x1080p (16:9) 25 fps HEVC(h.265). I am not an expert so I use Google AI for the ffmpeg preset.

C:\Users\pc_name\Desktop\Video_apps\ffmpeg-win64\bin\ffmpeg -i C:\Users\pc_name\Desktop\00010.MTS -vf "bwdif=mode=0arity=0" -c:v libx265 -crf 23 -preset medium -tag:v hvc1 -c:a copy -sn C:\Users\pc_name\Desktop\output.mp4

Some how that stupid sign () is messing up the preset string. Just replace it with : and p. Please refer to his image.

[Attachment 93284 - Click to enlarge]

You may need to change the directory settings.
ffmpeg = C:\Users\pc_name\Desktop\Video_apps\ffmpeg-win64\bin\
source file = C:\Users\pc_name\Desktop\
output file = C:\Users\pc_name\Desktop\

If you need to speed things up, then you can ask Google AI for QSV(Intel CPU) and Nvidia(GPU) for hardware acceleration preset.

I feel this is a cleaner approach. But I could be wrong.
BTW, I notice that youtube is using AVC 1920x1080p. Just launch google in AI mode and post the above preset and request for 1920x1080p in AVC format output.

[Attachment 93285 - Click to enlarge]

[Attachment 93286 - Click to enlarge]

[dellsam34]

What's the final output format?
You should be able to open the original HDV file with an Avisynth script in VirtualDub2 or AvsPmod and re-encode and de-interlace/resize in one step, without the need for an an intermediate lossless file.

AVSPmod would not open the original mpeg-2 file, It doesn't tell me the reason why.

[dellsam34]

Just sharing what I have done to de-interlace a video clip taken from a Sony DSC-HX20V camera in 1440x1080i (16:9) 25fps AVC(h.264) to 1440x1080p (16:9) 25 fps HEVC(h.265). I am not an expert so I use Google AI for the ffmpeg preset.

My HDV files are tape based, so they are in mpeg-2 not h.264.

[Sharc]

What's the final output format?
You should be able to open the original HDV file with an Avisynth script in VirtualDub2 or AvsPmod and re-encode and de-interlace/resize in one step, without the need for an an intermediate lossless file.

AVSPmod would not open the original mpeg-2 file, It doesn't tell me the reason why.

Did you try to open it in AVSPmod with different source filters (LWLibavVideoSource(), BSSource, ffms2 ...)?
Or if you have an Avisynth script you could open and run it in Vdub2 and skip AVSPmod.

[lollo]

Dellsam34, the procedure is correct and is what we have been promoting for years for YouTube uploads. Just few remarks:

- as noticed by others, no need to generate an intermediate large uncompressed file, especially with a useless 4:2:2 colorspace conversion (that later requires migrating back to 4:2:0)

- the preset "fast" for QTGMC may be not the optimal for quality (and I understand you are aiming at maximum quality)

- you can then load (follow Sharc suggetions) the AviSynth script with QTGMC and nnedi3 and the rest in VirtualDub to generate the final lossless file to be loaded in YT, or you canf feed the script to ffmpeg configuring h264 at low CRF if you prefere a small (but compressed) file to upload, with marginal degradation in the final video. Apparently you are not sensitive to big file creation so I would use the first

- eventually consider upscaling to higer resolution to take benefit of the latest YT algorithms evolution

[Alwyn]

Libav opens HDV after the usual flashing shenanigans, FFmpegSource2 opens it almost instantly; audio is in-sync.

Eg FFmpegSource2("F:\Family Videos\2009\20090713 Ferris Wheel\FW.mpg", atrack=-1)

[The 14th Doctor]

- eventually consider upscaling to higher resolution to take benefit of the latest YT algorithms evolution

Will agree on that front, upscaling to at least 1440p so that your finalized YouTube upload using the VP9 codec. This is why I upscale all my content to 4K, even Blu-ray rips.

[Sharc]

Libav opens HDV after the usual flashing shenanigans, ....

Some time is needed to produce the index file. So please be patient

[lollo]

Will agree on that front, upscaling to at least 1440p so that your finalized YouTube upload using the VP9 codec. This is why I upscale all my content to 4K, even Blu-ray rips.

BR ok, but be careful with SD to do not introduce artifacts. I am not a big fan of upscale, but I do it the few times I upload to YouTube.

[pchan]

Just sharing what I have done to de-interlace a video clip taken from a Sony DSC-HX20V camera in 1440x1080i (16:9) 25fps AVC(h.264) to 1440x1080p (16:9) 25 fps HEVC(h.265). I am not an expert so I use Google AI for the ffmpeg preset.

My HDV files are tape based, so they are in mpeg-2 not h.264.

Assuming that you can import from tape to PC in HDV format then you can encode to the format that Youtube preferred. Youtube re-encodes all videos uploaded anyway. Use mediainfo(you can download from Videohelp) and tell Google AI your video format and request AI to give you the ffmpeg preset for Youtube video. If you encounter ffmpeg error, cut & paste it and ask AI for fix.

[rgr]

I'm trying to work on some HDV files to put on YT so they don't get screwed up by YT de-interlacing algorithm, I'm taking the long way trying to preserve the quality, I want to know if there are quicker methods that result in less quality degradation, Here is what I'm doing so far:

- Input the 4:2:0 1080i HDV file into vdub2 to strip the mpeg-2 codec saving as 4:2:2 HuffYUV AVI (not sure if 10bit can make a difference here, so I'm just using 8bit).
- Input the resulting large file into AVSpmod and de-interlace with QTGMC in fast mode, followed by a resize in the second line from 1440x1080 to 1920x1080 to preserve the 16:9 aspect ratio.
- Upload the second resulting file into YT and let it do the encoding.
- Delete all the files but the original HDV source.

No need for intermediate file:

1. Create an AVS script:
BSsource("input.mpeg2")
ConverttoYUV422(bits=10, quality=true, interlaced=true)
QTGMC(preset="slow")
Lanczos4Resize(1920,1080)
Lsfplus(preset="slow")
PropSet("_SARNum", 1). PropSet("_SARDen", 1)
ConverttoRGB(quality=true).ConverttoYUV420(bits=8, matrix="709:l", quality=true)
Prefetch

2. Check the AVS script in VDub.

3. Compress the AVS file using FFmpeg:
ffmpeg.exe -AviSynth_flags all -i input.avs your_settings output.mp4

For better quality you can use Converttoyuv444 instead of YUV422.

[Sharc]

^^^ Chroma upsampling/downsampling with 3 conversions (4:2:0 -> 4:2:2 -> RGB -> 4:2:0) on what purpose/benefit?

[rgr]

1. 422P10 - for more accurate processing when more filters are used (personally I would prefer YUV444P16 considering that later there will be resize and each pixel will "weigh" more)
2. RGB->420 - converting to BT709 in AviSynth unfortunately requires going through RGB. With that bit depth, it doesn't matter.

However, there is an error here, it should look like this:
ConverttoRGB(quality=true,bits=16).ConverttoYUV420 (bits=8, matrix="709:l", quality=true)

[Sharc]

2. RGB->420 - converting to BT709 in AviSynth unfortunately requires going through RGB. With that bit depth, it doesn't matter.

Isn't the OP's 1080i HDV already 709? So no detour via RGB needed. I would leave it all in 4:2:0 anyway - and set the field parity before QTGMC() to prevent a surprise.

[video_fan]

2. RGB->420 - converting to BT709 in AviSynth unfortunately requires going through RGB. With that bit depth, it doesn't matter.

ColorMatrix("Rec.601->Rec.709") does not require any RGB conversion!

[cbehr91]

I would agree on upscaling slightly to 1440p for YouTube uploads and use CAS() to slightly sharpen the footage after upscaling. I've also had to use MosquitoNR at very low settings to remove MPEG compression noise on HDV footage.

[rgr]

2. RGB->420 - converting to BT709 in AviSynth unfortunately requires going through RGB. With that bit depth, it doesn't matter.

ColorMatrix("Rec.601->Rec.709") does not require any RGB conversion!

Yes, but ConvertToXXX with quality=true now operates in 32bit.
And I haven't heard good things about ColorMatrix.
But of course you can also use.

[video_fan]

2. RGB->420 - converting to BT709 in AviSynth unfortunately requires going through RGB. With that bit depth, it doesn't matter.

ColorMatrix("Rec.601->Rec.709") does not require any RGB conversion!

Yes, but ConvertToXXX with quality=true now operates in 32bit.
And I haven't heard good things about ColorMatrix.
But of course you can also use.

YUV <-> RGB is a lossy translation, so introducing it because wanting to work in 32bit is completely crazy

[VWestlife]

If you're going to upscale 1080i/1080p, you might as well just double it to 2160p.

1440p makes sense for 480i/480p footage (3X upscale) or 720p footage (2X upscale).

[dellsam34]

Did you try to open it in AVSPmod with different source filters (LWLibavVideoSource(), BSSource, ffms2 ...)?
Or if you have an Avisynth script you could open and run it in Vdub2 and skip AVSPmod.

LWLibavVideoSource did the trick, Thanks

[dellsam34]

If you're going to upscale 1080i/1080p, you might as well just double it to 2160p.

1440p makes sense for 480i/480p footage (3X upscale) or 720p footage (2X upscale).

Will consider that in the next file, Thanks.

[Sharc]

2. RGB->420 - converting to BT709 in AviSynth unfortunately requires going through RGB. With that bit depth, it doesn't matter.

ColorMatrix("Rec.601->Rec.709") does not require any RGB conversion!

Yes, but ConvertToXXX with quality=true now operates in 32bit.
And I haven't heard good things about ColorMatrix.
But of course you can also use.

Sidenote only: I usually prefer Avsresize (zlib) for such conversions.
http://avisynth.nl/index.php/Avsresize

[rgr]

2. RGB->420 - converting to BT709 in AviSynth unfortunately requires going through RGB. With that bit depth, it doesn't matter.

ColorMatrix("Rec.601->Rec.709") does not require any RGB conversion!

Yes, but ConvertToXXX with quality=true now operates in 32bit.
And I haven't heard good things about ColorMatrix.
But of course you can also use.

Sidenote only: I usually prefer Avsresize (zlib) for such conversions.
http://avisynth.nl/index.php/Avsresize

I used z_ConvertFormat once and saw that the colors did not match perfectly with those before the conversion (601->709)
Then I realized that zscale had severe bugs (and Avsresize is based on the same library), so it only confirmed to me that it was better not to use it.
https://code.ffmpeg.org/FFmpeg/FFmpeg/issues/23026

[rgr]

2. RGB->420 - converting to BT709 in AviSynth unfortunately requires going through RGB. With that bit depth, it doesn't matter.

ColorMatrix("Rec.601->Rec.709") does not require any RGB conversion!

Yes, but ConvertToXXX with quality=true now operates in 32bit.
And I haven't heard good things about ColorMatrix.
But of course you can also use.

YUV <-> RGB is a lossy translation, so introducing it because wanting to work in 32bit is completely crazy

For now, it is the most accurate. And yes, I'm still waiting for ConvertToYUV to enable accurate conversion without RGB. I guess it was supposed to work in current beta.

[video_fan]

https://code.ffmpeg.org/FFmpeg/FFmpeg/issues/23026

What? That bug is for changing color space. Here we can stay all way to 4:2:0

For now, it is the most accurate.

You must be fool, Chroma upsampling/downsampling with 3 conversions (4:2:0 -> 4:2:2 -> RGB -> 4:2:0) most accurate than ColorMatrix("Rec.601->Rec.709") ???

Sorry to disturb but this is completely crazy

[Sharc]

^^^ This. And where/why does the 601 sneak in at all? I assume the OP's 1080i HDV files (post#1) were already in 709 to begin with .... so just leave it alone, no?

[rgr]

https://code.ffmpeg.org/FFmpeg/FFmpeg/issues/23026

What? That bug is for changing color space. Here we can stay all way to 4:2:0

Staying within the narrowest possible range is the simplest and stupidest method to generate rounding errors using multiple filters.

For now, it is the most accurate.

You must be fool, Chroma upsampling/downsampling with 3 conversions (4:2:0 -> 4:2:2 -> RGB -> 4:2:0) most accurate than ColorMatrix("Rec.601->Rec.709") ???

Math isn't your strong suit, is it?

[rgr]

^^^ This. And where/why does the 601 sneak in at all? I assume the OP's 1080i HDV files (post#1) were already in 709 to begin with .... so just leave it alone, no?

And are there?