Chvrches - 2021-08-27 Masonic Lodge, Hollywood Forever Cemetery, Los Angeles, CA (PRO SBD TVEAI 1080p48)

Chvrches - 2021-08-27 Masonic Lodge, Hollywood Forever Cemetery, Los Angeles, CA (PRO SBD TVEAI 1080p48)

Chvrches
Masonic Lodge
Hollywood Forever Cemetery
Los Angeles, California
USA
August 27, 2021
2021-08-27

This is the first Screen Violence tour concert. It has been brightened and the logo removed. Detail has been enhanced, there's less compression, no judder, and framerate has been doubled from 24 to 48fps.

File size: 10800MB
Overall bit rate: 22800Kbps
Video: AVC1 [email protected] (H264) 1920x1080p 16:9 48.000fps 22000Kbps VBR CFR 195702 frames
Audio: AAC LC 44.1KHz 2 channel 126Kbps CBR (as FLAC, see below)
Total time: 01:07:57

Setlist:
He Said She Said
Forever
Leave a Trace
California
Bury It
Good Girls
Miracle
Science-Visions
How Not To Drown
Recover
The Killing Moon (live debut)
Never Say Die
The Mother We Share
Clearest Blue







Technical notes:
Similar to creating a movie or music album, lots of experiments were tried and trash canned. Nearly every Topaz VEAI model was tried and only one gave good results. FilmConvert film grain looked bad due to the dark lighting and strobes (I was attempting to reduce banding).

This is the output of a modified VEAI model I call "Dione Killer Upscale" that excels at creating video with sharp detailed features with minimal compression artifacts. You can't buy the model as it's my own creation based on another Topaz model (I've posted the how-to on Topaz forums).

Two versions of this were previously released, one on Dime at 3000Kbps and one on Sharemania at 6000Kbps. The Sharemania version has a few seconds cut at the beginning and end. The remaster adds the extra seconds from the Dime version. The audio on both releases is the same bitrate and quality and I used the Dime version. Both were released at 30fps with judder as the native rate is 24fps.

My first thought with the releases was that they were too dark and generally poor looking due to the overcompression and judder. Bringing up the levels was a challenge to not blowout scenes (that took about 20 encodes to get right). The tools are such (they suck) that you have to take it through the whole process and finally to your player to really see the effects of a change. I did use a short clip to minimize the number of iterations with the full video.

Artemis HQ and MQ v9 didn't do much. I tried Artemis Dehalo and Strong Dehalo expecting crap and that's what they delivered. I'm learning to not bother with Proteus. I had just developed the Dione Killer Upscale model, and that worked very well. This is the first release using it.

Getting 48fps was a major challenge and took a while. The issues are the darkness and strobe-like lighting. FrameRateConverter is my goto tool for difficult material, but it puked with this, delivering mostly blended frames. SVPFlow did the same. Flowframes and RIFE-app had severe artifacts. Davinci Resolve Studio's converter gave blended frames.

At this point there was one option left- VEIA's Chronos AI model for framerate conversion, which I had no experience with. The quality was excellent when it worked, but it was generating thousands of extra blended frames. I tried for a few weeks to find a solution that worked. I eventually found one comment that said it worked fine at 2x (I was trying to go from 24 to 50fps). I tried that and finally had success. This is an extremely compute intensive AI model, using about 25x more floating point calculations than FRC, taking 2.5 days on a single GTX 1080Ti GPU or 30 hours on 2 1080Ti GPUs.

I used the recently developed DPID for downsizing. This resizer uses all the data from supersampling to 200% scale to create a more detailed 1080p (standard resizers throw away this detail). There is some controversy over whether it works better than conventional scaling methods, but the theory behind it makes sense IMO.

The audio for How Not To Drown is not great. I fixed it as best as I could using a tube compressor, a hard limiter, and an AI based vocal isolator called DeepRemix (based on Spleeter). The issue is that the original audio was possibly screwed by some improper settings on several compressors and it sounded like they had a battle with the song. Locate the original audio if you want to hear how bad it got. I have released this with FLAC audio to avoid lossy re-encoding the original 126Kbps AAC LC audio.

Lineage Dime: webcast -> Streamlink -> FFmpeg (edit) -> uncompressed .ts
Video: MPEG4 Video (H264) 1920x1080p 16:9 3125 Kbps 30.00 fps
Audio: AAC LC 126 Kbps 44.1 KHz stereo

Lineage Sharemania: webcast -> ? -> .ts
Video: MPEG4 Video (H264) 1920x1080p 16:9 6000 Kbps 30.00 fps
Audio: AAC LC 126 Kbps 44.1 KHz stereo

Lineage video remaster: DGDecodeNV demux of both streams -> Virtualdub2 + Avisynth+ -> Lagarith .avi for each stream -> TVEAI 2.3 with Dione Killer Upscale model at 200% on each stream to .png sequence output -> created AutoIt scripts to rename the pngs appropriately to merge the two streams, the higher quality Sharemania is used for most of the video -> Virtualdub2 + Avisynth + Imagesource, DPID smart resizer, Tweak to make brighter, Inpaint for logo removal, outputting Lagarith .avi for VEAI -> VEAI 2.3 Chronos at 2x to get 48fps, .png output -> Virtualdub2 + Avisynth + Imagesource + x264 CRF 15.0 -> .mp4
Whew!

Lineage How Not To Drown remaster: argh...

Finally:
MKVMerge .mp4 + .flac -> .mkv

For the AV geeks, here are most of the details:
Create the .dgi files using DGIndexNV. I used Virtualdub2 with Avisynth+ for the scripts. The following is for the video only. I processed the audio separately.

Convert both videos to 24fps .avi videos:

01-SM.avs
DGSource("chvhfc-SM.dgi", deinterlace=0)
TFM()
TDecimate()

Save as Lagarith .avi.
Process with VEAI 2.3 using Dione Killer Upscale at 200%. Save as .png sequence.

01-DM.avs
DGSource("chvhfc-DM.dgi", deinterlace=0)
TFM()
TDecimate()

Save as Lagarith .avi.
Process with VEAI 2.3 using Dione Killer Upscale at 200%. Save as .png sequence (to a different folder).

There are many way to do this. This is the method I used: Combine the two videos (use the SM one for most of the video) via Windows Explorer, Irfanview and a renumbering program (I used 2 scripts running with Autoit for renumbering).

Resize, fix levels, remove logo. logo.bmp mask created with Videoredo's frame grabber and Photoshop:

02-run2-SM+DM-chvhfc-imgsrc-resize-inpaint.avs
SetFilterMTMode("DEFAULT_MT_MODE", MT_MULTI_INSTANCE)
SetFilterMTMode("ImageSource", MT_SERIALIZED)
SetFilterMTMode("InpaintFunc", MT_SERIALIZED)
Import("InpaintFunc.avs")
ImageSource("u:\wm\run2-SM+DM\%06d.png",fps=24,start=000000,end=097851,use_DevIL=true,pixel_type="RGB24")
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)
InpaintFunc(last,mask="u:\wm\logo.bmp",loc="X1",mode="Inpaint",speed=5,PP=75,PPmode=1)
prefetch(8)

Save as Lagarith .avi.
Process with VEAI 2.3 Chronos. Set to 48fps output. Save as .png sequence.

02-run2-SM+DM-chvhfc-run2-48fps.avs
ImageSource("v:\wm\run2-48\%06d.tif",fps=48,start=000000,end=195701,use_DevIL=true,pixel_type="RGB24")

Save as .mp4. I used x264 with Veryslow Film L4.2 CRF15.0. Set colorspace to Rec. 709. Component range Limited (16-235).


Mediainfo:
General
Unique ID : 225194486770676351710575978220355064260 (0xA96AE16AF484C06DBDEE91BAB75811C4)
Complete name : Chvrches - 2021-08-27 Masonic Lodge, Hollywood Forever Cemetery, Los Angeles, CA (PRO SBD TVEAI 1080p48).mkv
Format : Matroska
Format version : Version 4
File size : 10.8 GiB
Duration : 1 h 7 min
Overall bit rate mode : Variable
Overall bit rate : 22.8 Mb/s
Encoded date : UTC 2022-08-12 04:06:12
Writing application : mkvmerge v9.6.0 ('Slave To Your Mind') 32bit
Writing library : libebml v1.3.4 + libmatroska v1.4.5

Video
ID : 1
Format : AVC
Format/Info : Advanced Video Codec
Format profile : [email protected]
Format settings : CABAC / 4 Ref Frames
Format settings, CABAC : Yes
Format settings, Reference frames : 4 frames
Codec ID : V_MPEG4/ISO/AVC
Duration : 1 h 7 min
Bit rate : 22.0 Mb/s
Width : 1 920 pixels
Height : 1 080 pixels
Display aspect ratio : 16:9
Frame rate mode : Constant
Frame rate : 48.000 FPS
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Progressive
Bits/(Pixel*Frame) : 0.221
Stream size : 10.5 GiB (97%)
Writing library : x264 core 157 r2935M 545de2f
Encoding settings : cabac=1 / ref=4 / deblock=1:-1:-1 / analyse=0x3:0x133 / me=umh / subme=10 / psy=1 / psy_rd=1.00:0.15 / mixed_ref=1 / me_range=24 / chroma_me=1 / trellis=2 / 8x8dct=1 / cqm=0 / deadzone=21,11 / fast_pskip=1 / chroma_qp_offset=-3 / threads=12 / lookahead_threads=2 / sliced_threads=0 / nr=0 / decimate=1 / interlaced=0 / bluray_compat=0 / constrained_intra=0 / bframes=8 / b_pyramid=2 / b_adapt=2 / b_bias=0 / direct=3 / weightb=1 / open_gop=0 / weightp=2 / keyint=250 / keyint_min=25 / scenecut=40 / intra_refresh=0 / rc_lookahead=60 / rc=crf / mbtree=1 / crf=15.0 / qcomp=0.60 / qpmin=0 / qpmax=69 / qpstep=4 / ip_ratio=1.40 / aq=1:1.00
Default : Yes
Forced : No
Color range : Limited
Matrix coefficients : BT.709

Audio
ID : 2
Format : FLAC
Format/Info : Free Lossless Audio Codec
Codec ID : A_FLAC
Duration : 1 h 7 min
Bit rate mode : Variable
Bit rate : 783 kb/s
Channel(s) : 2 channels
Channel layout : L R
Sampling rate : 44.1 kHz
Frame rate : 10.767 FPS (4096 SPF)
Bit depth : 16 bits
Compression mode : Lossless
Delay relative to video : 77 ms
Stream size : 381 MiB (3%)
Writing library : libFLAC 1.2.1 (UTC 2007-09-17)
Language : English
Default : Yes
Forced : No