← Home ← Back to /g/

Thread 106341777

28 posts 28 images /g/
Anonymous No.106341777 >>106341811 >>106341901 >>106342600 >>106342675 >>106344367 >>106344381
Alright so since we don't know when sharty will enable jpeg xl on 4chan I started looking for alternatives and it surprised me that one was hiding right in the libjxl folder: jpegli, which practically speaking is the biggest threat to webp right now because it can make backwards compatible JPG images that have lower file sizes which mozjpeg couldn't do. As far as I understand JPG images encoded from say the 90s required around 2-3 BPP for high quality while those encoded with jpegli now only require 1-2 BPP so the file size reductions are pretty huge even for personal use.

Anyway for those interested I made a windows batch script that not only lets you select what quality level you want but also gives you a ssimulacra2 score and if the score isn't good enough, you can retry again without launching the script again. There's a lot of disagreement on what this should be but I've personally found diminishing returns around 85 so that's what I'm going to stick with.

@echo off
:loop
echo:
set /p q="Por favor quality: "
FOR /R %%G IN (*.png) DO cjpegli "%%~G" "%%~dpnG.jpg" ^
-q %q% --chroma_subsampling=444 -p 0 --fixed_code &&^
echo: && echo SSIMU2 = &&^
ssimulacra2 "%%~G" "%%~dpnG.jpg" && echo:
set /p retry="Test again? YES or NO?: "
if /i "%retry%"=="YES" goto loop
if /i "%retry%"=="yes" goto loop
if /i "%retry%"=="Yes" goto loop

Another really interesting thing I almost forgot to point out is one of the big reasons why most people hate webp is the fact that it only supports 4:2:0 chroma sub-sampling which IMHO is like a huge step backwards compared to JPG. Literally 75% of all color information is permanently destroyed when you use 4:2:0.
Anonymous No.106341811
>>106341777 (OP)

and how this color information was generated say with bayer image sensor
Anonymous No.106341869
Here's a complex test PNG image, notice how the soil has a lot of noisy detail. ~2.3 megapixels which would be ~300 KB at 1 BPP
Anonymous No.106341883 >>106341972 >>106342344
And here is the high quality jpegli version with a ssimulacra2 score of ~85.3 and at ~600 KB file size is ~2 BPP
Anonymous No.106341901 >>106341938 >>106346373
>>106341777 (OP)
Nice virus, pal
Anonymous No.106341938
>>106341901
You COULD punch this into the numerous free AI chatbots on the web and ask what my script aimed at people in MIDDLE SCHOOL does...

But I guess trolling is less effort than that, huh?
Anonymous No.106341972
>>106341883

thats european space program people look tracktor motormonoblock on auction like holy artefact and it is on earth
Anonymous No.106342226 >>106342431
I want jpeg to die so badly, bros.
Anonymous No.106342344
>>106341883
I reencoded your png with gimp default settings and got a jpg of roughly the same size as this one, but noticeably worse quality (on the right). Pretty cool
Anonymous No.106342431
>>106342226

you hate 8x8 compression blocks do you?
Anonymous No.106342600 >>106342673 >>106342743
>>106341777 (OP)
It really breathed new life into the format and has one neat trick up its sleeve.
Anonymous No.106342634 >>106342743
Anonymous No.106342657 >>106342743
Anonymous No.106342673 >>106342720
>>106342600

but these are all same 0.8 curve after level matching?

you really would want curve to pop like 1.1
Anonymous No.106342675
>>106341777 (OP)
lurk moar zoomer
Anonymous No.106342720
>>106342673
I didn't build the charts it's from Gianni Rosato's site, the SVT-AV1-PSY dev
Anonymous No.106342743
>>106342600
>>106342634
>>106342657
As cool as these are I tested xyb jpegli images on multiple devices like smart TVs and old Android tablets and they all failed to decode. Still a nice little preview of what jpeg xl offers.
Anonymous No.106343769
OP here I've decided to post a few meme worthy images remastered in SSIMU2 ~85 via the jpegli encoder. I'll also use photoshop's denoising to clean them up if the best source I find is a dirty JPG. PNG version first and the the jpegli one next, maybe these can be used for image codec testing in the future.

Anyway here's the first one. 0.3 megapixels/~40 KB @ 1 BPP
Anonymous No.106343778
Por favor quality: 86
Read 574x514 image, 173404 bytes.
Encoding [YUV444 d1.000 AQ p0 FIX]
Compressed to 57773 bytes (1.567 bpp).
574 x 514, 56.873 MP/s [56.87, 56.87], , 1 reps, 1 threads.

SSIMU2 =
85.18997432
Anonymous No.106344188 >>106344326 >>106344519
How many pixels can jxl fit into 1-2kb?
Anonymous No.106344326
>>106344188
Just a curiosity because 1600 bytes is about the max you can paste into 4chan as base85 text.
Anonymous No.106344367
>>106341777 (OP)
gib bash script with linux-compatible tools
Anonymous No.106344381 >>106344438
>>106341777 (OP)
>jpegli
>webp
both jewgle cancer
Anonymous No.106344438
>>106344381
So is jpeg-xl you dumbass. What, would you stop shitting in a toilet if google had invented the toilet as well?

Google's Role in JPEG XL Development
Co-creation: JPEG XL was the result of a collaboration between Google and Cloudinary.
PIK Format: Google submitted its PIK image format to the JPEG XL Call for Proposals in 2018, and it was a fundamental component of the new codec.
Joint Development: After PIK and FUIF were selected to be combined, Google and Cloudinary engaged in several years of joint development to refine the format.
Anonymous No.106344519
>>106344188

maybe 128x128
Anonymous No.106344711
Dam I miss may mays like this. Now it's just twitter screen shit and jaks these days.

0.4 megapixels/~50 KB @ 1 BPP
Anonymous No.106344714
Por favor quality: 86
Read 620x620 image, 416664 bytes.
Encoding [YUV444 d1.000 AQ p0 FIX]
Compressed to 64521 bytes (1.343 bpp).
620 x 620, 63.714 MP/s [63.71, 63.71], , 1 reps, 1 threads.

SSIMU2 =
85.52133760
Anonymous No.106346373
>>106341901
sure there is