Scribd
Upload
834 views2 pages

Encoded by JoyBell (UTR)

This document discusses various encoding settings and quality metrics used by JoyBell when encoding video files, including: low quality (LQ) fixed bitrate encoding, 10-bit encoding, Nvidia NVENC encoding at quality levels Q18 and Q22, the improved quality of Nvidia Turing NVENC compared to Pascal, and the use of SSIM quality scores (ranging from 0-100) to indicate how closely an encoded file matches the source material. The document also describes JoyBell's "Tuned Encode" process which combines the best settings from 2-pass and Q22 encodes to achieve higher quality.

Uploaded by

toza vitaminoza
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
834 views2 pages

Encoded by JoyBell (UTR)

This document discusses various encoding settings and quality metrics used by JoyBell when encoding video files, including: low quality (LQ) fixed bitrate encoding, 10-bit encoding, Nvidia NVENC encoding at quality levels Q18 and Q22, the improved quality of Nvidia Turing NVENC compared to Pascal, and the use of SSIM quality scores (ranging from 0-100) to indicate how closely an encoded file matches the source material. The document also describes JoyBell's "Tuned Encode" process which combines the best settings from 2-pass and Q22 encodes to achieve higher quality.

Uploaded by

toza vitaminoza
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
You are on page 1/ 2

This was proudly encoded by JoyBell using x265/HEVC and AAC audio.

I am a member of Unity Team Release group, remember to search "UTR".

My torrents are posted on our own website: UNITEAM.CO


as well as extratorrent and 1337x and ETTV.

"What the difference between lq, 10bit, NVEnc, q18 and q22?"
The LQ version is fixed bitrate two pass of variable image quality.
The ones with 10bit and not a Q number are just 2 pass slow encodes.
The Q22 (CRF22) version is fixed quality of variable size.
The Q18 (CRF18) version is fixed blu-ray quality of variable size.
The NV or NVEnc version is using the Turing Nvenc encoder for either Q22 or 2 pass.
UpRez or AI version is my work upsampling and sharpening and attempts at increasing
detail. UpRez is with passes of traditional filters, which are very slow. AI using
modern inference AI software that takes hudreds or thousands of hours to produce a
single movie worth of content.

NVidia's Turing Video Encoder is much improved over the previous Pascal versions.
My internal testing at lossless showed this size for same image ratios:
Software x265 3.0 = x1.00 Filesize
NVenc Pascal = x2.34 Filesize
NVenc Turing = x1.16 Filesize

That is for Turing encode to match the quality of the software encode it would need
to be 16% larger.

IF the Q22 version is smaller than the 10bit version, that means that the video
was easy to compress to that quality.
AND in that case the 10bit version is higher quality OR You can just look at my
quality score I give you!

Tuned Encode Tag


The Tuned Encode is a full effort by me. First, I take samples and take every
encoder setting one at a time finding the best result at a reasonable performance
no slower than half the speed of slow. I combine the settings with a measurable
quality improvement into a specific encoding setting for that content.
Then I combine the best of my 2 pass and q22 encodes.
I use some math to take the quality difference of the two encodes plus a kick based
off the S score under 100 to try and get the best quality. Usually this ends up in
a file size between the two pass and q22 but at higher quality than either.

"What does the S## or FS## on some of your titles mean?"


We are experimenting with using a SSIM based quality metric for our encodes to help
users understand the quality of our work. The SSIM measures the accuracy of the
outputted encode verse the source. It does not reflect on the sharpness or
awesomeness of the image; only how close it is to the source material. An "S##" is
a score where filters where not used. "FS##" is an encode where we used filters
before the encoder, usually for nasty amounts of grain, this can throw off the
score usually in a positive way and we like to be clear and honest at UTR. I also
have been using a sophisticated sharpen filter, this can make a much more detailed
and sharper image, but this has the effect of lowering the S score. To be clear the
SSIM score represents how accurately we encoded the file, a crappy looking movie
from the 90's with an FS89 will probably still look worse than a current UHD movie
with a S45.
Some content, usually lacking detail doesn't score very high no matter what the
bitrate. I have a q18 encode at 25,000KBs that only scored S64. For us trying to
get the best quality for you, that still lets us compare settings for that file,
higher is still better.
To make things simple I think of it as a grade; where a S91 would be an A- and a
S78 would be a C effort.

Anyway, try not to be fooled by the dumpers taking credit for our work.

Love you all,


JoyBell

You might also like