Skip to main content ↑ ↓
Topic: jepgli vs webp - how difficult is the mathematics? (Read 289 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

jepgli vs webp - how difficult is the mathematics?

I'm blowing up the steam. You may or may not like it:

I'm writing this, but without pleasure. Its not my duty to teach mathematics. Peaple should have already finish their school. And every school has  math courses.

Recently I've noticed  a lot of articles about this "jpegli" "new encoder" from google. There is even a long discussion on ycombinator. I would like to ask you (as experienced programmers), what do you think about those idiots?

Almost all articles and posts talk about "dethroning and getting rid of webp aka jpegli is better than webp". NOT!. Simply its the problem of jpeg (actually jfif). Mathematically it is very bad. It is what it is.

Furthermore its not a "new" codec. Its still mathematically the same type part of jfif encoder. Its not even developed as new encoder, but some quantization tables were experimentally put  into jpg from jxl project (some sub-github of jxl) Since I couldn't find it initially. Its under cjxl on github and under jxl in the Artix repo. There is no such thing as jpegli.

This "new jpeg"  are just some quantization tables, thrown onto whatever jpeg encoder.  Since the guy learned something by creating some cjxl encoder (he's a bad programmer...), he put some tables into some jpgeg. Tha's it.  These tables.... suck! They may even be worse than any other jpeg. But inherently this "new" codec is still ... very bad. Because mathematically its still jpeg. End of story.

There is more. You cannot compress 10-bit png with jpegli (will not convert to 8-bit). It does not work. Just look at the result. Oh, my! Oh, and this new color space. Total fail. Unreadable file (horizontal lines...). In the end its still jpg. And very bad jpeg. Combine both and you have bad resulst. How can it beat vp9? Cannot. And cwebp is very well written. VERY GOOD CODING (just compare it to cjxl and look at low ram options in cwebp or looseless compression in cwebp*. CWEBP guys win. CJXL- back to school!). And it cannot be 10-bit (people at ycombinator made this up...). This "new color space" is not even backwards compatible.


Just talke a look. I went extra mile to show it to you. Look at the screenshots, where I wentextra mile to test the "new jpeg compression from goolge that will rid of webp".

For all those internet "programmers": STOP WRITING ARTICLES AND REPEAT BULLSHIT YOU MATHEMATICAL MORONS. I hope people here know better. I can only hope that, I will not have to wear a T-shirt: "I'm with stupid" , while eating some deep-fried corndog on a stick.


I am very sorry for the runt, but I have no time to/nor do I want to - register/post on those reddits/phoronixes etc. Not my cup of tea to explain anything to idiots.

Just take a look at the attachment. Seeing is believing. Even people who studied art will understand those screenshot.

webp vs jepgli compression artifacts (bunch of screenshots):


https://upload.adminforge.de/r/tH7ChK6a5k#9MPrvbyO126Jzy50Ts/wmNcVLodF6LUGZav3fP/bqOM=

P.S. I used the best options (man pages).

* compresses versy well and is decoded extremly fast with loseless encoding. cjxl losess takes ages to encode and decode (!) . Day and night. CJXL - back to school