Guetzli: Google Rolls Out A New JPEG Encoder

[mczak]

As a person living in Switzerland, I find the naming "Guetzli" for this algorhythm deeply embarassing.

FWIW I find it pretty funny. And the name is hardly a surprise even, google has compression algorithms for other purposes already named zopfli and brotli...

[TheLexMachine]

If we could get this working with Open CL and Intel's QuickSync JPEG encoding hardware, it would be a big boon to Linux users.

[liam]

The paper was interesting, especially this:

We reduce JPEG artifacts by capturing images at the highest JPEG
quality level using a Canon EOS 600d camera and downsampling the re-
sulting images by 4x4 using Lanczos resampling, as implemented in GIMP.
Photographers often apply unsharp masking to compensate for downsam-
pling, so we also apply it in most of the images before downsampling. The
degree of unsharp masking is chosen arbitrarily, but before any compression
experiments.

That seems completely unnecessary and adds some uncertainty, to me, about their results.

They admit that the butteraugli pv model is designed for high bit rate (they targeted 2.6bpp for the test, which correspond to libjpeg settings 83-93, for their image set), visually lossless archival/long-term storage.
I haven't looked at the pv model they used (frankly, I'd not heard of it before) but I'd be surprised if there was a serious serialization issue inherent to it. The developers seem more interested in getting out their results and playing with the new toy, so it's not surprising that the code is slow.

Also, https://github.com/mozilla/mozjpeg/issues/242

[rabcor]

High quality JPEGs? Funniest joke I've heard all year!

[rgloor]

As a person living in Switzerland, I find the naming "Guetzli" for this algorhythm deeply embarassing.

I'd rather find it funny. Especially because I could imagine, that people in some regions of the world might have some challenges in trying to pronounce it. LOL.

If you consider that that google moved some developmemt to switzerland purely for tax reasons, that makes it all the more embarrassing. Sorry for being Swiss is all that comes to my mind.

I think, this did play a minor part, if at all. There are a lot of countries around the world, offering much better tax advantages. And as far as I know, a high standard of living and an ideal environment for a good work-life-balance was most important, in order to attract the best possible employees.


But if somebody does not like a certain corporation, for whatever reason, that is another story.

[slalomsk8er]

I'd rather find it funny. Especially because I could imagine, that people in some regions of the world might have some challenges in trying to pronounce it. LOL.

Me too, as a Swiss person I find the name funny. Why can a simple thing like the name of a piece of software embarrass you? How do you feel about master and slave processes?

I feel rater disturbed by the SJW victim culture that likes to create hurt feelings by constructing context in purely technical topics that is not there. There tendency to shame other people in to changing there ways is some thing that makes me very angry.

Back on topic. Why a new encoder? JPEG is create once serve often and as such it totally makes sense to invest upfront to gain later in faster transmission and lower bandwidth use.

[linuxgeex]

Hopefully Timo Kokkonen incorporates this into jpegoptim and we can all get the benefits without even caring where it came from or what it's called ;-)

Frack. Just went to the github project page and saw this:

Guetzli is a JPEG encoder that aims for excellent compression density at high visual quality. Guetzli-generated images are typically 20-30% smaller than images of equivalent quality generated by libjpeg. Guetzli generates only sequential (nonprogressive) JPEGs due to faster decompression speeds they offer.

Well, that removes it from my use cases. The only thing I use JPEG for is mobile website media, and progressive loading mitigates slow loading on marginal 2G/3G (what phones fall back to when they can't get LTE/4G)

mozjpeg's cjpeg supports progressive. I find -quality 76 -optimize -progressive -dct float -smooth 100 produces a very good web default image. The perceptual quality between 75 and 76 is striking for gradients, and the file size barely moves. -smooth 100 does a great job at fixing dithering noise introduced by digital cameras and some gif-sourced images, but it can eat the definition of tiny text and fine line art... both of which really should be compressed with PNG, or kept in a vector format (web font, SVG.) If you're running a CMS and users are uploading JPEG files with such art in them then you might want to turn it off since educating them is probably more expensive than the bandwidth, lol.

[Toboe]

[...]
I feel rater disturbed by the SJW victim culture that likes to create hurt feelings by constructing context in purely technical topics that is not there. There tendency to shame other people in to changing there ways is some thing that makes me very angry.
[...]

And we found the person who finds a way to shoehorn SJW into it, congratulations!

[dado023]

Using FileOptimizer9.60 to convert gueztli jpg to progressive gives even more savings on file size.

[liam]

High quality JPEGs? Funniest joke I've heard all year!

They can be excellent, but i find this test to be extremely weird (as i explained above). What i don't understand is why
I think it was Monty from xiph who called it (iirc) "spooky, alien technology". Significantly improving upon it in every case has been really difficult. I think a reason for this is that the spec is supposed to be excellent (according to monty).

https://blog.mozilla.org/research/20...h-mozjpeg-2-0/