processing notes for Guns N' Roses 2023-06-24 Glastonbury 2023 Festival, Pilton, UK (PRO 1080p50)

[xavier242]

Reference:
Guns N' Roses 2023-06-23 Glastonbury Festival, Worthy Farm, Pilton, UK (PRO_HD)

This will be a locked work-in-process thread and I'll unlock it when I'm done (for questions and comments).

There is a strong belief that any re-encoding (re-compression) of video is bad. This is incorrect. The answer is that it depends on several variables.

Know that FEED videos occasionally appear on the internet. They have been re-encoded 3 or more times already. These are high bitrate, high quality streams that the broadcast and webcast industries use to create the resolutions and bitrates they send to viewers.

I have yet to see an excellent 1080p 50fps version of the Guns N' Roses Glastonbury 2023 gig. The DARKFLIX release is not great. There's not much detail and it's about 4x the size it should be for that level of quality. Excellent would be clear, sharp, 50fps video. As it's almost 2-1/2 hours, about 20GB would be required for 1080p50.

23.976fps is great for movies, dramas, sitcoms, scifi as the low framerate adds to the cinematic effect and pulls you into the narrative. For live performances and sports, a higher framerate and bitrate helps bring you to the venue (IMO).

You need to start with the best quality you can find. None of the tools including Topaz VEAI can do much if the source is lacking detail. In this case, the 24.9GB 3840x2160 video posted to bittorrent has the best quality and sufficient detail. I'd say it's a FEED stream capture, but the bitrate is low for the resolution (unless the industry has lowered their standards). Bits/(Pixel*Frame) is only 0.058 (barely adequate for H265 4K video).

Mediainfo says 50fps but that's been known to be inaccurate with interlaced video. A lot of 50fps content is really 25fps with each frame repeated once (fake 50fps). Real video at 25i coming from a videocamera is 50 interlaced fields per second shifted one pixel up and down, alternating with each field.

If you open this up in a good editor like VideoRedo, it will show each field. This is the first step as the rest of the process depends on what you find. You need to determine if you have fake 50fps or real 50fps.

...to be continued

After analysis with VRD it's actually 3840x2160p 50fps (progressive video), so the above is "for reference" should you come across an interlaced HD video.

[xavier242]

process notes:
guns_n_roses-glastonbury_(2023)-ddc-hdr-2160p-h265-2023-srp.mkv
rename:
gnrg2023.mkv

v01
HEVC 10bit 4:2:0 (Type 2) 24900MB 3840x2160p 24200Kbps 02:27:32.19 50fps 16:9 Bits/(Pixel*Frame)=0.058 AAC LC 192Kbps 48KHz
442618 max frame num (0 is a frame)
442619 frames total
This is likely real video- 50 unique frames per second from the cameras. Modern cameras can do 50p, not limited to 25i anymore (pros don't typically use post-interpolation as it takes too long)

Demux with gMKVExtractGUI 2.1.0.0 (MKVToolNix 9.6.0) to make:
gnrg2023_track1_und.hevc
gnrg2023_track2_eng_DELAY 12ms.aac

Run DGIndexNV.exe with gnrg2023_track1_und.hevc to create gnrg2023_track1_und.dgi.

This is a rare case where only one Avisynth script file was needed (when using TVEAI, 2 are needed for pre and post processing). .avs script files are created with any text editor (Notepad).

Note the use of DPID to downsize. This is an intelligent resizer that doesn't toss detail when downscaling (unlike the others). It uses a crapload of CPU. Multi-instance is a must.

01.avs
SetFilterMTMode("DEFAULT_MT_MODE", MT_MULTI_INSTANCE)
SetFilterMTMode("DGSource", MT_SERIALIZED)
DGSource("gnrg2023_track1_und.dgi", deinterlace=0)
#ConvertToYV12(interlaced=false)
DPID(1920,1080,lambdaU=1.7,lambdaV=1.7,lambdaY=1.7)
#Tweak(bright=-3,cont=1.23,sat=1.00,hue=0)
#CAS(sharpness=0.6)
#vsMSharpen(threshold=6.0, strength=30.0, mask=false, luma=true, chroma=true)
prefetch(8)

In Virtualdub2 x264 8bit dialog:
x264 8bit Slower Film High L4.2 YUV 4:2:0 SAR 1/1 CRF 17.4

Make sure to set Pixel Format:
Color space: Rec. 709 (HD)
Component range: Limited (Y:16-235)

Save video as MP4 (MPEG-4 Part 14):
01-gnrg2023-DPID1.7-crf17.4.mp4

Tweak, CAS, and vsMSharpen were not used (the "#" marks a comment). They are sometimes needed when working with TVEAI and DVD video.

It's compressing now at about 12fps.. I'll see how it looks tomorrow.

[xavier242]

It's done and looks very good. It's sharp without applying any traditional sharpening. DPID is an amazing scaler when downsizing, producing an AI like detail enhancement. The video is 19.7GB and Bits/(Pixel*Frame) is 0.185.

the rest of the process:
Mux the original audio back in (gnrg2023_track2_eng_DELAY 12ms.aac).
I used MKVToolNik GUI 9.60 (this version was just before some major bugs were introduced; probably fixed now but I don't upgrade for upgrading's sake). Be sure to set Default duration to 50p and change the Default track flags to yes. Also set the audio language to English.

One point to note, the video was created with HDR.. which I don't like but changing that is days or weeks of tweaking (with Tweak) since I've not tried removing HDR before. I'll have to search and see if anyone created an un-HDR script. Anyone have experience with unHDRing?

The original 4K video would probably respond well to a TVEAI enhancement (staying at 4K, but sharper). I need to build a dual 3090 box.. but money is kind of tight at the moment. Perhaps next year.