|
Jyrki Alakuijala
|
2021-04-07 02:01:16
|
deadline for new features is tomorrow -- we know if we made it on Friday
|
|
2021-04-07 02:01:49
|
otherwise it will be six weeks later -- but possible to use canary builds in between if one is really eager
|
|
|
Jim
|
2021-04-07 02:27:46
|
The upcoming version 89 of Firefox is scheduled to enable AVIF by default (though Safari and most mobile do not yet support). So you could go that route... it just won't look as good. <:kekw:808717074305122316>
|
|
|
Scope
|
2021-04-07 02:35:46
|
<https://www.reddit.com/r/jpegxl/comments/mm3dt7/merged_into_chromium_implement_imagejxl_decoding/>
|
|
|
|
Jyrki Alakuijala
|
2021-04-07 03:43:44
|
looks like we made it
|
|
|
_wb_
|
2021-04-07 04:07:47
|
So jxl-behind-a-flag will be in Chrome M91?
|
|
2021-04-07 04:09:51
|
first non-canary release will be end of May, then?
|
|
2021-04-07 04:10:26
|
and we can try to get default-on in M92, stable release end of July?
|
|
2021-04-07 04:10:53
|
https://chromiumdash.appspot.com/schedule I don't quite understand what all these phases are...
|
|
|
eddie.zato
|
|
Crixis
|
|
eddie.zato
|
|
2021-04-07 04:23:59
|
Where is the flag?
|
|
|
|
eddie.zato
|
2021-04-07 04:24:56
|
Open 'about:flags' and search for JXL
|
|
|
|
Crixis
|
2021-04-07 04:24:58
|
a no, I'm Version 91.0.4469.2
|
|
2021-04-07 04:26:23
|
How i have lower version with canary?
|
|
|
|
Deleted User
|
2021-04-07 04:28:23
|
You need Chromium, not Google Chrome (Canary)
|
|
|
|
eddie.zato
|
2021-04-07 04:28:24
|
The latest Chromium builds https://download-chromium.appspot.com/
|
|
|
|
Crixis
|
2021-04-07 04:29:09
|
Ah lol chromium not chrome
|
|
2021-04-07 04:30:20
|
I'll wait 2 or 3 days
|
|
|
|
Scope
|
2021-04-07 04:31:58
|
https://mo271.github.io/jxl/
|
|
|
|
Deleted User
|
2021-04-07 04:44:48
|
Has anyone else tried decoding images with noise synthesis? Mine crash Chromium tab...
|
|
|
|
veluca
|
2021-04-07 04:49:29
|
that's weird...
|
|
|
fab
|
2021-04-07 05:14:03
|
i can't set jpeg xl to open in chromium
|
|
2021-04-07 05:14:06
|
how to do
|
|
2021-04-07 05:14:15
|
windows fails
|
|
2021-04-07 05:15:58
|
it doesn't do
|
|
2021-04-07 05:16:11
|
even by the file association tool
|
|
2021-04-07 05:16:20
|
it says to change program
|
|
2021-04-07 05:20:04
|
file:///C:/Users/User/Documents/gallery/s.png.jxl
|
|
2021-04-07 05:20:08
|
it doesn't open anything
|
|
2021-04-07 05:20:15
|
and the jxl is enabled by default
|
|
2021-04-07 05:20:49
|
i click enabled
|
|
2021-04-07 05:21:18
|
|
|
|
|
Deleted User
|
2021-04-07 05:25:30
|
<@!416586441058025472> go to voice channel and make screen sharing, I'll try to help you
|
|
|
|
fab
|
2021-04-07 05:30:13
|
maybe is windows 7
|
|
2021-04-07 05:30:17
|
that is bugged
|
|
2021-04-07 05:30:30
|
that can't use too much applications
|
|
|
|
_wb_
|
2021-04-07 05:30:47
|
Could be that file:// doesn't work, as then it doesn't get the media type?
|
|
|
|
fab
|
2021-04-07 05:30:51
|
then how to force windows 7
|
|
2021-04-07 05:30:53
|
file works
|
|
|
|
_wb_
|
2021-04-07 05:31:05
|
Ah
|
|
2021-04-07 05:31:35
|
You want to open jxl files by default with your browser? Why not with a viewer?
|
|
|
|
fab
|
2021-04-07 05:31:48
|
is way more secure
|
|
2021-04-07 05:31:52
|
to use a viewer
|
|
2021-04-07 05:32:06
|
yes but
|
|
2021-04-07 05:32:21
|
imageglass has too ugly bars it looks old in 2021
|
|
2021-04-07 05:33:08
|
like in ffmpeg it would get integrated
|
|
|
raysar
|
2021-04-07 05:33:38
|
Yes i don't know how to keep xnview mp in ram, starting is so slow. I need to ask it in xnview forum.
|
|
|
|
fab
|
2021-04-07 05:34:11
|
and i t doesnt' open the image but a small preview with other images
|
|
2021-04-07 05:35:59
|
is there dark mode for xnview
|
|
2021-04-07 05:38:44
|
i can't use dark mode is bugged it displays white in full screen and black in normal mode
|
|
2021-04-07 05:39:17
|
and also the theme selection you need to have normal xnview
|
|
2021-04-07 05:39:49
|
do you think is xnview the problem or that my computer hasn't the latest RAM modules
|
|
|
|
raysar
|
2021-04-07 05:42:03
|
i'm testing decoding jxl is slower with chromium than xnviewmp, it's clear on big resolution pictures.
|
|
|
|
Scope
|
2021-04-07 05:47:42
|
Yes, especially on large lossless images, I think it's from the reason that browsers use single-threaded decoding (since images are mostly more than one)
|
|
|
|
fab
|
2021-04-07 05:48:48
|
maybe chromium hasn't the author
|
|
2021-04-07 05:48:56
|
is only an exe without author and firm
|
|
2021-04-07 05:49:08
|
so what should i do to force chromium to open jxl
|
|
2021-04-07 05:49:13
|
without copying the path
|
|
2021-04-07 05:49:29
|
i'm on windows 7
|
|
2021-04-07 05:49:44
|
how to force windows to open jxl in chromium
|
|
2021-04-07 05:49:49
|
local files
|
|
|
Dr. Taco
|
2021-04-07 07:11:40
|
why is this picture so tall, https://mo271.github.io/assets/images/jpg.jpg
|
|
2021-04-07 07:12:17
|
and why is the text so aliased
|
|
|
|
Deleted User
|
2021-04-07 07:16:14
|
I've just found another glitch, but with JPEG XL, not Chrome. Try combining `-m`, `--lossy-palette` and `-Q <quality>` (lossy Modular with lossy palette).
|
|
|
|
Scope
|
2021-04-07 07:23:37
|
Yep, `--lossy-palette` is incompatible and does not work with lossy modular from the beginning (this is not really a bug, except that this combination is allowed in the options)
|
|
|
|
Dr. Taco
|
|
spider-mario
|
2021-04-07 07:29:40
|
one possible reason for chrome as a default image viewer is that it has at least some sort of color management, even if it’s flawed
|
|
2021-04-07 07:29:49
|
a surprising number of image viewers don’t have any color management
|
|
2021-04-07 07:30:14
|
for example, Windows Photos completely ignores the monitor profile
|
|
2021-04-07 07:30:25
|
unlike the viewer from Windows XP which took it into account correctly…
|
|
2021-04-07 07:32:39
|
or maybe it was starting in Vista
|
|
2021-04-07 07:36:41
|
(https://community.adobe.com/t5/lightroom-classic/photos-highly-saturated-when-exported-from-lightroom/m-p/9943541#M111962, https://forum.luminous-landscape.com/index.php?topic=117152.20)
|
|
2021-04-07 07:36:44
|
(sorry for the rant)
|
|
|
username
|
2021-04-07 10:41:27
|
interesting to see how the final decoded image looks in chromium compared to the WIC codec
|
|
|
|
spider-mario
|
2021-04-07 11:10:21
|
is that Windows 7, or is that somehow a 7-like theme for a more recent Windows version?
|
|
|
|
username
|
|
spider-mario
is that Windows 7, or is that somehow a 7-like theme for a more recent Windows version?
|
|
2021-04-07 11:30:04
|
windows 8.1 with aero glass, a windows vista theme and some other stuff I sprinkled on top
|
|
|
|
Jim
|
2021-04-08 12:07:17
|
Looks like the photo viewer might be using an older version
|
|
|
|
username
|
2021-04-08 12:19:56
|
it is
|
|
2021-04-08 12:20:07
|
i'm using the WIC codec from here https://github.com/mirillis/jpegxl-wic
|
|
2021-04-08 12:22:05
|
which is from early february
|
|
|
|
il1kesonic
|
2021-04-08 01:09:51
|
guys a made an emoji 🥺 🥵
|
|
|
doajc_blogger
|
2021-04-08 01:13:23
|
I'm sorry for inviting her. I didn't know she would do this here.
|
|
|
BlueSwordM
|
2021-04-08 01:13:38
|
<@&803357352664891472> I think it's time to restrict her image permissions
https://discord.com/channels/794206087879852103/794206170445119489/829524124485419039
|
|
|
|
il1kesonic
|
2021-04-08 01:14:18
|
yall rlly some obese ass discord mods
|
|
2021-04-08 01:14:46
|
|
|
|
190n
|
2021-04-08 01:18:37
|
**^^^ volume warning**
|
|
|
|
doajc_blogger
|
2021-04-08 01:20:09
|
I'm in a DM call with her and tried to persuade her to stop but she refuses so you might have to ban her.
|
|
|
|
il1kesonic
|
2021-04-08 01:21:15
|
i didnt refuse i deleted the message :angy:
|
|
|
Moritz Firsching
|
|
Dr. Taco
why is this picture so tall, https://mo271.github.io/assets/images/jpg.jpg
|
|
2021-04-08 07:12:43
|
The reason why jpg.jpg is so tall is that I used an image of the same size as the jxl.jxl image, which also has the tall JXL logo. The looks so bad because that I quickly used gimp to generate the images without taking great care in making them pretty. Feel free to make a pull request to change those files here: https://github.com/mo271/mo271.github.io (I guess one could also make this page a bit nicer, center the images etc.. )
|
|
|
Has anyone else tried decoding images with noise synthesis? Mine crash Chromium tab...
|
|
2021-04-08 07:13:44
|
I would be interested in any specific image which makes chrome crash, together with the information on what system that happens
|
|
|
|
fab
|
2021-04-08 07:48:32
|
i want a better jxl viewer, a gimp editor
support for using the photo as new tab
for using as a wallpaper
for using in android
|
|
2021-04-08 07:49:14
|
also updated jpeg xl wic with information if the photo is lossy modular lossy vardct or lossless or n1 or palette
|
|
|
|
spider-mario
|
2021-04-08 08:27:17
|
the first four lines almost sound like song lyrics
|
|
|
Petr
|
2021-04-08 08:29:53
|
And who knows, maybe not only visual jxl-art but also music jxl-art could help spreading the format. 😄
|
|
|
|
fab
|
2021-04-08 08:51:26
|
like at q 72.6 it just compressed
|
|
2021-04-08 08:54:34
|
those are in av1 server before 8 august
|
|
2021-04-08 08:54:46
|
|
|
2021-04-08 08:54:50
|
but you can't decode those files with current viewer or decoders
|
|
2021-04-08 08:54:56
|
and the quality is much lower
|
|
2021-04-08 08:55:21
|
|
|
2021-04-08 08:55:23
|
i remember i did 72.6 with also random enhancement of daala, cavif rs didn't existed at that time
|
|
2021-04-08 08:55:49
|
to post images of instagram
|
|
2021-04-08 08:56:10
|
now i tried to restore the quality with compression to have a similar quality to now
|
|
2021-04-08 08:56:16
|
i edited gamma etc
|
|
2021-04-08 08:56:58
|
|
|
2021-04-08 08:57:34
|
for most images i did only the latest command, for others i had to make more changes
|
|
2021-04-08 08:58:07
|
the images i had from instagram were compressed maybe with https://discord.com/channels/794206087879852103/794206170445119489/829640550948405258 at q 72.6 modular
|
|
2021-04-08 08:58:34
|
or vardct (worse i don't have the courage to try)
|
|
|
Jake Archibald
|
2021-04-08 09:17:29
|
I think I've found a bug, but I'm not sure where to file it:
1. Take the 1920x version of https://unsplash.com/photos/idP6ct9jkmk
2. Pop it into https://squoosh.app
3. JPEG XL, 75 quality, progressive rendering, effort 4
4. Output is 184kB
5. Change effort to 5
6. Output is 405kB
This issue seems to depend on 'progressive rendering'
|
|
2021-04-08 09:25:55
|
Same issue with effort=7
|
|
|
|
Crixis
|
2021-04-08 09:32:14
|
i can replicate on cjxl 0.3.7
|
|
2021-04-08 09:32:28
|
-s 8
|
|
2021-04-08 09:32:53
|
-s 8 --progressive
|
|
2021-04-08 09:32:57
|
|
|
|
|
Jyrki Alakuijala
|
2021-04-08 09:48:15
|
we are not very responsive today since we are trying to solve a memory regression in Chrome
|
|
2021-04-08 09:48:49
|
this bug is interesting and we will look at it next week
|
|
|
|
veluca
|
2021-04-08 09:50:08
|
but please file it on gitlab
|
|
|
|
Jake Archibald
|
2021-04-08 10:05:36
|
<@!179701849576833024> as in https://gitlab.com/groups/wg1/-/issues? If so, which project?
|
|
|
|
veluca
|
2021-04-08 10:06:05
|
https://gitlab.com/wg1/jpeg-xl/-/issues
|
|
|
|
Jake Archibald
|
2021-04-08 10:06:18
|
ta
|
|
2021-04-08 10:11:58
|
Filed https://gitlab.com/wg1/jpeg-xl/-/issues/204. No rush btw! Good luck with the memory issue
|
|
|
|
Jyrki Alakuijala
|
2021-04-08 10:45:31
|
the memory issue is mystical -- it is like when you fill up a car and the windshield wipers start moving automatically
|
|
|
|
_wb_
|
2021-04-08 10:47:04
|
mystical issues are usually things that turn out to be silly bugs that make you wonder "why did this ever work in the first place?"
|
|
|
|
Jim
|
2021-04-08 10:48:50
|
The wipers are like a dog's tail. The car is just happy you are giving it a drink.
|
|
|
|
raysar
|
2021-04-08 11:10:23
|
i learn that multitreading is only enable for < s7. s8 and s9 don't have this feature. You need to explain it in the --help.
|
|
|
|
Jim
|
2021-04-08 11:36:09
|
Interesting. That would explain the amount of time it takes to run the slowest speeds. I just figured it was looking at so many different things it took it a long time to do. I assume the code has not been threaded yet or maybe something in there can't be threaded?
|
|
|
|
Crixis
|
|
raysar
i learn that multitreading is only enable for < s7. s8 and s9 don't have this feature. You need to explain it in the --help.
|
|
2021-04-08 11:58:12
|
I don't think this is exact
|
|
|
|
_wb_
|
2021-04-08 12:01:54
|
There's a big difference between modular and vardct
|
|
|
|
fab
|
2021-04-08 12:04:53
|
what this does
|
|
2021-04-08 12:04:54
|
https://gitlab.com/wg1/jpeg-xl/-/commit/ef50d4f7f0cfaab312a18ac59b10b19b292e322c
|
|
2021-04-08 12:10:06
|
|
|
2021-04-08 12:10:45
|
does this contains all images or is the right code
|
|
2021-04-08 12:10:53
|
i saved with chromium
|
|
|
sklwmp
|
|
raysar
i learn that multitreading is only enable for < s7. s8 and s9 don't have this feature. You need to explain it in the --help.
|
|
2021-04-08 12:13:51
|
Are s8 and s9 still considered useful or will they be deprecated sometime? I remember someone making a Reddit comment (I forgot where) that speed 8 and 9 may not even be consistently better in the future than speed 7, just consistently *slower*.
|
|
2021-04-08 12:13:59
|
Sorry if I misunderstood that, by the way
|
|
2021-04-08 12:15:16
|
I still use s9 just because I have a lot of time on my hands and it seems to compress better
|
|
2021-04-08 12:16:33
|
I also add `-E 3 -I 1` to my CLI options, but I'm not exactly sure what they do, though. Could someone explain what they do and why they could possibly make the file size smaller?
|
|
|
|
fab
|
2021-04-08 12:17:32
|
This maximum compression for using lossless
|
|
2021-04-08 12:17:33
|
for %i in (C:\Users\User\Documents\m3\*.png) do cjxl "%i" "%i.jxl" -q 100 -m -s 9 -E 3 --num_threads=2
|
|
2021-04-08 12:17:42
|
jpeg is not supported in 0.3.7
|
|
2021-04-08 12:17:46
|
before it was
|
|
2021-04-08 12:18:33
|
in next version maybe you could use jpg as input and just specify -s 7 or -s 9 --num_threads=2
|
|
2021-04-08 12:18:56
|
and it will do lossless jpg transcode with VARDCT (brunsli was deprecated)
|
|
2021-04-08 12:19:08
|
more info on wikipedia
|
|
2021-04-08 12:19:31
|
or on documentation of the reference encoder, there are also experimental man pages
|
|
2021-04-08 12:20:26
|
about using speed 8 and speed 9, speed 9 now produces good results for lossy provided the images is really RA
|
|
2021-04-08 12:30:24
|
read the manpages in 0.3.7
|
|
|
|
Crixis
|
|
sklwmp
Are s8 and s9 still considered useful or will they be deprecated sometime? I remember someone making a Reddit comment (I forgot where) that speed 8 and 9 may not even be consistently better in the future than speed 7, just consistently *slower*.
|
|
2021-04-08 12:33:28
|
in Vardct s7 is the safes bet, in modular mode s 8 and s9 are better then s 7
|
|
|
sklwmp
I also add `-E 3 -I 1` to my CLI options, but I'm not exactly sure what they do, though. Could someone explain what they do and why they could possibly make the file size smaller?
|
|
2021-04-08 12:35:40
|
-E 3 is more then s 9
|
|
|
|
fab
|
2021-04-08 12:37:39
|
modular s 8 is very versatile in jxl
|
|
2021-04-08 12:37:49
|
just don't use lower modular lossy
|
|
2021-04-08 12:38:05
|
and lossless start from s 7
|
|
2021-04-08 12:38:24
|
i recomme
|
|
2021-04-08 12:38:46
|
lower modular lossy just squeeze the pixels without perceptual metrics
|
|
2021-04-08 12:39:24
|
not recommended unless you know what you doing and you do at least at 86.6 q
|
|
2021-04-08 12:40:38
|
certains images with jxl can go from 16 mb to 4,9 mb in lossless
|
|
2021-04-08 12:40:53
|
but not with jxl lossless transcode
|
|
2021-04-08 12:42:03
|
jxl lossless transcode achives 22%, you can do more with paq200px, paq compressors, but is a sequential compressor, decodes and encodes too slow and at same speed, so images needs quick display and you can't and the format hasn't a version or a specific bitstream
|
|
2021-04-08 12:42:30
|
so it can't be read on browsers and it uses too much RAM, especially cmix
|
|
2021-04-08 12:42:48
|
but every the specific one like EMMA, BMF, don't look for them
|
|
|
|
sklwmp
|
|
Crixis
-E 3 is more then s 9
|
|
2021-04-08 12:42:50
|
wdym?
|
|
|
|
fab
|
2021-04-08 12:43:17
|
maybe it means extensive 3
|
|
2021-04-08 12:43:33
|
it does more like 9,1 mb becomes 8,2 mb
|
|
2021-04-08 12:43:38
|
or so
|
|
|
|
sklwmp
|
2021-04-08 12:43:46
|
```
-E K, --extra-properties=K
[modular encoding] number of extra MA tree properties to use```
I found this after adding some `-v` flags to `cjxl --help`
|
|
|
|
Crixis
|
|
sklwmp
wdym?
|
|
2021-04-08 12:44:10
|
-s 9 -E 3 is slower than -s 9 but more efficient, is a sort of -s 10
|
|
|
|
fab
|
2021-04-08 12:44:16
|
https://docs.google.com/spreadsheets/d/1ju4q1WkaXT7WoxZINmQpf4ElgMD2VMlqeDN2DuZ6yJ8/edit#gid=2120588660
|
|
2021-04-08 12:44:39
|
do not use webp2 it deletes invisible pixels that are near alpha background
|
|
2021-04-08 12:44:53
|
it compresses more like on emoji but it isn't very useful
|
|
|
|
Crixis
|
|
sklwmp
```
-E K, --extra-properties=K
[modular encoding] number of extra MA tree properties to use```
I found this after adding some `-v` flags to `cjxl --help`
|
|
2021-04-08 12:44:55
|
yep, more features than s9
|
|
|
|
fab
|
2021-04-08 12:47:09
|
but i suggest speed 7 if you have a good cpu like that can do more than 2000 geekbench
|
|
|
|
Crixis
|
|
sklwmp
```
-E K, --extra-properties=K
[modular encoding] number of extra MA tree properties to use```
I found this after adding some `-v` flags to `cjxl --help`
|
|
2021-04-08 12:47:44
|
-h -v -v -v for all options
|
|
|
YAGPDB.xyz
|
2021-04-08 12:47:44
|
Couldn't find command '-v -v -v for all options'
|
|
|
|
fab
|
2021-04-08 12:47:50
|
speed 9 sometimes is fast for example a 720x720 file
|
|
|
|
Crixis
|
|
Couldn't find command '-v -v -v for all options'
|
|
2021-04-08 12:48:08
|
wut
|
|
|
|
Jim
|
2021-04-08 12:53:05
|
What do you mean by casual? That was discussed on here in the past and would be a great feature for many - such as graphic designers - who often do not know if they should use lossy or lossless.
|
|
|
|
sklwmp
|
2021-04-08 12:54:18
|
thanks for explaining `-E 3`, now what does `-I 1` do?
```
-I F, --iterations=F
[modular encoding] fraction of pixels used to learn MA trees (default=0.5, try 0 for no MA and fast decode)
```
I think I should really study up on how the compression works first
|
|
|
|
Crixis
|
|
Jim
What do you mean by casual? That was discussed on here in the past and would be a great feature for many - such as graphic designers - who often do not know if they should use lossy or lossless.
|
|
2021-04-08 12:55:48
|
there are yet some lossless in lossy vardct jxl, es. patches for text
|
|
|
sklwmp
thanks for explaining `-E 3`, now what does `-I 1` do?
```
-I F, --iterations=F
[modular encoding] fraction of pixels used to learn MA trees (default=0.5, try 0 for no MA and fast decode)
```
I think I should really study up on how the compression works first
|
|
2021-04-08 12:56:43
|
use all source pixel for some optimization (default 50% alias -I 0.5)
|
|
|
|
fab
|
|
sklwmp
thanks for explaining `-E 3`, now what does `-I 1` do?
```
-I F, --iterations=F
[modular encoding] fraction of pixels used to learn MA trees (default=0.5, try 0 for no MA and fast decode)
```
I think I should really study up on how the compression works first
|
|
2021-04-08 12:57:22
|
https://discord.com/channels/794206087879852103/794206170445119489/828717523385057300
|
|
2021-04-08 12:57:32
|
WebP compression without loss works thanks to a prediction technique where each pixel block is generated mathematically based on others that surround it.
|
|
2021-04-08 12:57:49
|
wikipedia jpeg xl english version is good enough to understand
|
|
|
|
Crixis
|
|
sklwmp
thanks for explaining `-E 3`, now what does `-I 1` do?
```
-I F, --iterations=F
[modular encoding] fraction of pixels used to learn MA trees (default=0.5, try 0 for no MA and fast decode)
```
I think I should really study up on how the compression works first
|
|
2021-04-08 12:58:16
|
also -g 3 is useful at times
|
|
|
|
fab
|
2021-04-08 12:58:18
|
even with loss
|
|
2021-04-08 12:58:26
|
it uses DCT
|
|
2021-04-08 12:58:44
|
vardct is just DCT with different block sizes
|
|
2021-04-08 12:59:01
|
the tools are written in wikipedia
|
|
2021-04-08 12:59:32
|
explanation -I (iterations): Now it is just fraction of data used for MA tree learning, where
I=100% is the maximum
|
|
2021-04-08 01:00:52
|
chromium works already for jpeg xl
|
|
2021-04-08 01:01:11
|
skip the new heuristic part it shouldn't be used as it causes a lot of blocking
|
|
|
|
Jim
|
|
Crixis
there are yet some lossless in lossy vardct jxl, es. patches for text
|
|
2021-04-08 01:09:00
|
I would suggest more along the lines of determining if the majority of the image appears to be vector graphics to use modular rather than vardct. There is a switchover where not having photographic data often results in a smaller file size using modular.
|
|
|
|
Crixis
|
|
Jim
I would suggest more along the lines of determining if the majority of the image appears to be vector graphics to use modular rather than vardct. There is a switchover where not having photographic data often results in a smaller file size using modular.
|
|
2021-04-08 01:12:29
|
for what i've read the future goal is use modular for all the parts of the image that require it
|
|
|
|
fab
|
2021-04-08 01:17:03
|
also i forget but facebook is interested in JXL
|
|
|
|
Jim
|
2021-04-08 01:17:09
|
That might be a tough one. Not so much if the background is a solid color but photographic data and where best the edge would be, but still fits.
|
|
|
|
fab
|
2021-04-08 01:17:42
|
and it uses an API of android called Facebook Fresco that can display images as older as android 2.1, similar to imagemagick for windo
|
|
2021-04-08 01:17:55
|
like you can add every format
|
|
2021-04-08 01:19:18
|
Jpeg XL is an encoder hahahaah
|
|
2021-04-08 01:19:30
|
false information
|
|
2021-04-08 01:19:42
|
is only a codec
|
|
2021-04-08 01:20:42
|
is a powerpoint
|
|
2021-04-08 01:21:07
|
<@!172952901075992586> do you have that link
|
|
|
|
Jim
|
2021-04-08 01:21:13
|
It's an encoder and decoder. No library would be able to display an image format without knowing how to read the binary data and convert to pixel info.
|
|
2021-04-08 01:21:17
|
Link to what?
|
|
|
|
fab
|
2021-04-08 01:22:25
|
https://www.youtube.com/watch?v=t63DBrQCUWc
|
|
2021-04-08 01:22:36
|
https://docs.google.com/presentation/d/1Q7ZgFC5ZPLWOo8PcQV1jfTfstMsOeisTlJvWcdUULEw/edit#slide=id.g8a04a98194_0_5
|
|
2021-04-08 01:28:13
|
this link is old
|
|
|
|
_wb_
|
2021-04-08 01:28:49
|
https://docs.google.com/presentation/d/1LlmUR0Uoh4dgT3DjanLjhlXrk_5W2nJBDqDAMbhe8v8/edit?usp=sharing
|
|
2021-04-08 01:28:59
|
those slides are slightly more up to date
|
|
|
|
fab
|
|
Crixis
use all source pixel for some optimization (default 50% alias -I 0.5)
|
|
2021-04-08 01:35:11
|
ah i didn't know
|
|
2021-04-08 01:35:26
|
where you read that?
|
|
|
|
Crixis
|
2021-04-08 01:44:35
|
in the help
|
|
|
|
fab
|
2021-04-08 01:46:06
|
manpages or gitlab files md
|
|
|
|
raysar
|
2021-04-08 02:22:04
|
It's in the roadmap to add jxl as input in cjxl encoder?
|
|
|
|
_wb_
|
2021-04-08 02:24:57
|
that's more for a `jxltran` kind of tool I think
|
|
|
|
Jim
|
2021-04-08 02:25:32
|
There is already a bug: https://gitlab.com/wg1/jpeg-xl/-/issues/103
|
|
|
|
il1kesonic
|
2021-04-08 02:29:22
|
daddy
|
|
|
|
_wb_
|
2021-04-08 02:29:23
|
it's trivial to just decode jxl and treat it as pixels, but a more interesting thing would be to do lossless recompression of lossy jxl too
|
|
|
|
Jim
|
2021-04-08 02:30:58
|
Not sure how that would even work - wouldn't you just end up with the same file?
|
|
|
|
_wb_
|
2021-04-08 02:33:04
|
there are some entropy coding choices that can be optimized given more time
|
|
2021-04-08 02:33:24
|
not a lot though
|
|
|
|
BlueSwordM
|
2021-04-08 02:33:33
|
For now, it'd just be nice if you could transcode lossless JXL to lossy.
|
|
2021-04-08 02:33:51
|
Would beat having a copy in PNG or having to decode manually. 😛
|
|
|
|
fab
|
2021-04-08 03:15:21
|
can E 1.6 be used
|
|
|
|
_wb_
|
2021-04-08 03:17:57
|
-E is how many previous channels to allow looking at in the context model
|
|
2021-04-08 03:19:14
|
So e.g. for RGBA without color transforms, -E 1 means G can 'see' R, B can see G, A can see B (they can look at 1 previous channel)
|
|
2021-04-08 03:19:50
|
-E 3 means G can see R, B can see R and G, A can see R, G and B.
|
|
2021-04-08 03:20:56
|
(after RCT it's more like chroma seeing luma but you get the idea)
|
|
2021-04-08 03:21:28
|
By default we do -E 0 because looking at previous channels comes at a speed cost in both encode and decode
|
|
2021-04-08 03:21:36
|
But it can be useful
|
|
|
|
fab
|
2021-04-08 03:22:39
|
error
|
|
|
|
raysar
|
2021-04-08 04:17:11
|
So i'm searching how to pipe in windows powershell image
`magick.exe C:\in.jxl - | cjxl.exe - .\out.jxl -d 2 -s 7`
did not works like linux
|
|
|
|
fab
|
2021-04-08 04:28:12
|
like this i would like
|
|
2021-04-08 04:28:13
|
for %i in (C:\Users\User\Documents\11\*.png) do cjxl3 "%i" "%i.jxl" -q 14 -s 4 --use_new_heuristics --faster_decoding=5 --num_threads=2
|
|
2021-04-08 04:28:34
|
and it automatically transform an image like that
|
|
2021-04-08 05:01:14
|
ok i'm trying newer builds
|
|
2021-04-08 05:01:16
|
for %i in (C:\Users\User\Documents\14\*.png) do cjxl4 "%i" "%i.jxl" -Y 12 -g 2 -X 8.8 -q 30.2 -s 4 -m --num_threads=2
for %i in (C:\Users\User\Documents\15\*.png) do cjxl4 "%i" "%i.jxl" -q 12.22 -s 8 --faster_decoding=7 --num_threads=2
|
|
2021-04-08 05:01:20
|
try to run
|
|
2021-04-08 05:03:53
|
works all
|
|
|
|
Deleted User
|
|
Moritz Firsching
I would be interested in any specific image which makes chrome crash, together with the information on what system that happens
|
|
2021-04-08 05:21:47
|
Only this file (found on Wikipedia, I added noise in GIMP) with noise synthesis decodes without issues. It's really small.
|
|
2021-04-08 05:25:22
|
But this one fails. (Chrome tab error: `STATUS_ACCESS_VIOLATION`)
|
|
2021-04-08 05:28:57
|
Another one, without noise synth, decodes without issues...
|
|
2021-04-08 05:29:32
|
...but its noise-estimated counterpart fails in Chrome with same error code.
|
|
2021-04-08 05:30:17
|
HDR images with noise are affected, too.
|
|
2021-04-08 05:31:36
|
Every .jxl file uploaded above should decode without problems with `djxl`. Only Chromium is going crazy.
|
|
|
|
fab
|
2021-04-08 05:45:36
|
|
|
2021-04-08 05:46:34
|
Now is a1248445
|
|
2021-04-08 05:46:57
|
like this type of extreme compression isn't impossible
|
|
2021-04-08 05:47:15
|
you just need to wait until they make improvements
|
|
2021-04-08 05:47:32
|
i don't know if is 0.4.0 or other commit/tag
|
|
|
|
Deleted User
|
2021-04-08 05:56:59
|
<@!795684063032901642> here's what the situation looks like with 2 of the files above, one with no noise synth and one with.
|
|
|
|
fab
|
|
Only this file (found on Wikipedia, I added noise in GIMP) with noise synthesis decodes without issues. It's really small.
|
|
2021-04-08 05:59:39
|
how you added noise in gimp
|
|
|
|
Deleted User
|
2021-04-08 06:00:11
|
Depends on your GIMP version. Which one do you have?
|
|
|
|
fab
|
2021-04-08 06:05:52
|
|
|
2021-04-08 06:06:02
|
a bit old
|
|
|
|
Deleted User
|
|
fab
a bit old
|
|
2021-04-08 06:19:24
|
Don't worry 😉
|
|
2021-04-08 06:20:05
|
Filters -> Noise -> ICH CIE Noise
|
|
|
|
_wb_
|
2021-04-08 06:45:11
|
|
|
2021-04-08 06:45:25
|
can confirm
|
|
2021-04-08 06:46:29
|
so what's the difference between the files? the ones that don't work are encoded with --noise ?
|
|
|
|
Deleted User
|
|
_wb_
can confirm
|
|
2021-04-08 06:46:55
|
That's with basically any file encoded with `--noise=1`. Only the smallest one (PNG-Gradient-noisy.jxl) managed to get decoded.
|
|
|
_wb_
so what's the difference between the files? the ones that don't work are encoded with --noise ?
|
|
2021-04-08 06:47:08
|
Yes, that's the only difference. You can see in the movie above.
|
|
|
|
_wb_
|
2021-04-08 06:48:00
|
a bit strange that djxl has no issue with it but in chromium it indeed doesn't work
|
|
2021-04-08 06:48:38
|
this is why things are behind a flag and considered experimental at first 🙂
|
|
|
|
Deleted User
|
2021-04-08 06:50:19
|
Can you open the smallest file I posted?
|
|
2021-04-08 06:51:08
|
Because my Chromium didn't crash on such a small file.
|
|
|
|
_wb_
|
2021-04-08 06:53:17
|
yes the smallest one is fine
|
|
2021-04-08 06:53:37
|
might be a bug that only surfaces in multi-group images with noise, or something
|
|
|
|
Deleted User
|
2021-04-08 06:54:32
|
The medium one (Paris_Night.jxl) is 400x480 and it's affected, too.
|
|
|
|
_wb_
|
2021-04-08 06:55:18
|
that's 4 groups though
|
|
2021-04-08 06:55:30
|
groups are 256x256
|
|
|
|
Deleted User
|
2021-04-08 06:55:37
|
Ah, ok
|
|
|
|
_wb_
|
2021-04-08 06:56:06
|
anyway, if it's easy to reproduce, it shouldn't be too hard to debug
|
|
|
|
fab
|
2021-04-08 07:00:08
|
|
|
2021-04-08 07:00:21
|
the preview (os windows 7)
|
|
2021-04-08 07:00:31
|
and https://github.com/saschanaz/jxl-winthumb
|
|
2021-04-08 07:00:37
|
doesn' t appear sharp
|
|
|
|
_wb_
|
2021-04-08 07:01:17
|
yuck
|
|
2021-04-08 07:01:28
|
I guess they're doing poor downscaling
|
|
|
|
Deleted User
|
|
_wb_
groups are 256x256
|
|
2021-04-08 07:05:57
|
Indeed, 256x256 file with noise synth gets decoded correctly
|
|
2021-04-08 07:10:12
|
But 257x257 file also decodes without issues... <:Thonk:805904896879493180>
|
|
|
|
veluca
|
2021-04-08 08:17:26
|
does it fail with decode_oneshot?
|
|
|
|
_wb_
|
2021-04-09 09:06:35
|
not for me
|
|
|
|
Moritz Firsching
|
2021-04-09 09:38:07
|
Thanks <@456226577798135808> for reporting the crashing images, we are looking into fixing this. It looks like this is not a problem with the chrome's jxl decoder, but we rather fix it upstream...
|
|
|
|
veluca
|
|
But 257x257 file also decodes without issues... <:Thonk:805904896879493180>
|
|
2021-04-09 10:42:51
|
FYI the problem threshold is at about 288 pixels 😄
|
|
|
|
_wb_
|
2021-04-09 10:52:28
|
Luca fixed it already
|
|
2021-04-09 01:58:17
|
slightly updated https://sneyers.info/jxl/
|
|
|
Master Of Zen
|
2021-04-09 02:04:15
|
<:PepeHands:808829977608323112>
|
|
|
|
_wb_
|
2021-04-09 02:14:50
|
|
|
2021-04-09 02:15:48
|
|
|
|
Nova Aurora
|
|
_wb_
|
|
2021-04-09 08:17:38
|
Why must firefox have so little money to throw at adopting every new standard <:PepeHands:808829977608323112>
|
|
|
|
Crixis
|
2021-04-09 08:20:12
|
Becouse firefox have no money at all
|
|
|
|
Nova Aurora
|
2021-04-09 08:22:29
|
Yeah Mozilla is still almost entirely reliant on Google <:PepeHands:808829977608323112>
|
|
2021-04-09 08:23:55
|
And now they've lost servo and rust.
|
|
2021-04-09 08:24:18
|
And I don't think any of their attempts to make more money currently will work
|
|
|
|
BlueSwordM
|
|
Nova Aurora
And I don't think any of their attempts to make more money currently will work
|
|
2021-04-09 08:24:43
|
Well, there is an easy way to make money for Mozilla.
|
|
2021-04-09 08:25:10
|
Basically, the US government should force Firefox to be preinstalled on Android phones rather than Chrome 😈
|
|
2021-04-09 08:25:42
|
Basically, let's see if Chrome won on mobile because it was better or because it was preinstalled.
|
|
|
|
Scope
|
|
|
Nova Aurora
|
2021-04-09 08:27:16
|
As a firefox mobile user, it was mostly cause chrome was preinstalled but also I don't think FF was really ready for mobile until the fenix update
|
|
|
BlueSwordM
Basically, the US government should force Firefox to be preinstalled on Android phones rather than Chrome 😈
|
|
2021-04-09 08:29:05
|
The government - both US and EU - want as much browser choice as possible, not one being forced on consumers
|
|
2021-04-09 08:34:10
|
I mean consumers don't seem to notice that every browser on IOS is rebranded Safari
|
|
|
|
_wb_
|
2021-04-09 09:14:32
|
I think the web platform may have become too complicated to have a healthy ecosystem with several different browser engines
|
|
2021-04-09 09:15:30
|
Making a good, up to date, fully featured and secure browser from scratch would take how many man years?
|
|
|
Pieter
|
2021-04-09 09:21:57
|
Starting where? With a graphics/text rendering engine? With a typical OS? Kernel? TCP/IP stack? Machine language? Silicon and the laws of physics?
|
|
|
|
_wb_
|
2021-04-09 09:31:02
|
Let's say starting with a typical OS.
|
|
|
|
Scope
|
2021-04-10 01:56:26
|
About using JXL in game engines and why textures need their own formats
<https://developer.unigine.com/forum/topic/7381-jpeg-xl-implementation/>
|
|
|
|
_wb_
|
2021-04-10 02:21:21
|
I wonder how much computational effort it is to compress RGBA to some gpu texture format
|
|
2021-04-10 02:22:12
|
Storing textures as jxl and decoding to a gpu format seems like a good idea
|
|
|
|
Scope
|
2021-04-10 02:36:53
|
I think it is possible, but most likely the formats originally designed for this will be more optimal, such as Basis <https://github.com/binomialLLC/basis_universal>
|
|
2021-04-10 02:40:43
|
However, games usually need images not only for textures (e.g. for UI, Art or viewing pictures inside the game, etc.)
|
|
|
Orum
|
2021-04-10 04:00:14
|
yeah, all the GPU formats are some constant ratio, which JXL is not
|
|
2021-04-10 04:02:33
|
Still, some games, and more than a few mods, provide lossless textures and let either the game or users recompress them if they desire better performance/less VRAM use. I suppose since it surpasses PNG & WebP it might see use there.
|
|
|
|
il1kesonic
|
|
|
spider-mario
|
2021-04-11 10:48:07
|
who and why?
|
|
|
|
fab
|
2021-04-11 11:51:22
|
delete this user
|
|
|
|
Nova Aurora
|
|
spider-mario
who and why?
|
|
2021-04-11 05:35:13
|
Sophia is a troll mostly
|
|
|
|
BlueSwordM
|
2021-04-11 05:35:47
|
Basically, I would restrict her image permissions.
|
|
2021-04-11 05:35:56
|
She posted some dubious content last week.
|
|
|
|
spider-mario
|
|
Nova Aurora
Sophia is a troll mostly
|
|
2021-04-11 05:36:00
|
that was my impression, but I thought it could be interesting to approach this with curiosity rather than hostility
|
|
|
|
Scope
|
2021-04-11 05:44:22
|
Mostly it happens when a person is completely uninterested in the topics of the server and communication often goes into trolling
|
|
|
|
username
|
2021-04-11 05:52:24
|
they seem to have ended up in this server not due to a interest in Jpeg XL or image formats but because they where invited to the server by someone who is their friend and the friend who invited even said that It may have not been the best idea and also if you look through their message history in this server you can see absolutely none of it is serious and none of it contributes to any conversation
|
|
2021-04-11 05:53:39
|
it might be best to kick them or something because they seem to not be interested in anything that is happening in here and are just posting random things and at some point it might lead to something not good happening
|
|
|
|
Deleted User
|
|
username
they seem to have ended up in this server not due to a interest in Jpeg XL or image formats but because they where invited to the server by someone who is their friend and the friend who invited even said that It may have not been the best idea and also if you look through their message history in this server you can see absolutely none of it is serious and none of it contributes to any conversation
|
|
2021-04-11 05:53:48
|
BTW what Discord client are you using? It looks unusual to me...
|
|
|
|
_wb_
|
2021-04-11 05:54:33
|
I will kick when it becomes too spammy
|
|
|
|
BlueSwordM
|
2021-04-11 05:55:08
|
Honestly, I don't get why people aren't interested in stuff like JPEG-XL.
|
|
2021-04-11 05:55:17
|
The technology, the workings, and math behind codecs in general is so interesting. <:BlobYay:806132268186861619>
|
|
|
|
fab
|
2021-04-11 05:59:17
|
doajc blogger posted a build of jpeg xl wic
|
|
|
|
username
|
|
BTW what Discord client are you using? It looks unusual to me...
|
|
2021-04-11 06:00:28
|
it's the normal client mostly (It's still the electron client discord provides at the base of it) but I have a theme applied to it and also some plugins that show me more information immediately without having to go dig through the ui to find
|
|
|
|
Deleted User
|
2021-04-11 06:00:51
|
How did you do that?
|
|
|
Scientia
|
2021-04-11 06:07:53
|
it reminds me a lot of the old black android UI on samsungs
|
|
|
|
username
|
2021-04-11 06:10:37
|
It's meant to try to replicate the sort of style UIs had around 2008
|
|
|
|
Scientia
|
2021-04-11 06:10:46
|
https://fdn.gsmarena.com/vv/reviewsimg/samsung-galaxy-nexus/sshots/gsmarena_003.jpg
|
|
2021-04-11 06:10:54
|
like this
|
|
2021-04-11 06:11:13
|
i remember i had a galaxy nexus
|
|
2021-04-11 06:11:43
|
sorry if this is slightly offtopic
|
|
|
|
raysar
|
2021-04-12 12:04:54
|
That's fun, with --target_size we can go so low in bpp 😄 look at this 1920*1440, 10kB
|
|
2021-04-12 12:07:52
|
at 0.06 bpp
|
|
2021-04-12 12:14:34
|
if i resize it at the same size, quality is so better.
I think we need to think about the limit of lowest quality. When we reach it, we need to resize the picture for a better compression.
In video there are study for that.
|
|
|
|
fab
|
2021-04-12 06:56:07
|
this doesn't load in sasha plug in
|
|
2021-04-12 06:56:23
|
i'm trying in nomacs
|
|
2021-04-12 06:58:43
|
what it changes between them
|
|
|
|
Crixis
|
|
raysar
if i resize it at the same size, quality is so better.
I think we need to think about the limit of lowest quality. When we reach it, we need to resize the picture for a better compression.
In video there are study for that.
|
|
2021-04-12 07:00:24
|
there is also a cjxl flag for resize
|
|
|
|
fab
|
2021-04-12 08:57:47
|
also battery will last less with jpeg xl
|
|
|
andentri
|
2021-04-12 07:41:13
|
Hey there.
Could someone be kind and explain why I encounter this error :
cjxl -s 9 -m -Q 85 --gaborish=1 -v --override_bitdepth 4 -c 0 -C 0 small.png small.jxl
djxl small.jxl small.png
[...]
`Read 930 compressed bytes [v0.3.7 | SIMD supported: SSE4,Scalar]
Decoded to pixels.
**Failed to write decoded image**.`
is there some inconsistency with the command line options ?
|
|
|
|
_wb_
|
2021-04-12 07:43:34
|
Try djxl --bits_per_sample 8
|
|
2021-04-12 07:43:55
|
Will do that by default after next sync
|
|
2021-04-12 07:44:25
|
Png writer doesn't know how to write 4-bit png, because we were too lazy to implement that code path
|
|
|
|
andentri
|
2021-04-12 07:45:12
|
Thx a lot (as always) !
|
|
2021-04-12 07:49:46
|
But..
It's strange ; with `djxl --bits_per_sample 8` I have a far less good color accuracy (ssimulacra and eyes proved)
|
|
|
|
Deleted User
|
2021-04-12 07:50:18
|
Could you send the original `small.png` file, please?
|
|
|
|
andentri
|
2021-04-12 07:53:24
|
|
|
2021-04-12 07:55:23
|
Or this one
|
|
2021-04-12 07:58:42
|
Original and success(0.49925809) + failed (0.48550824)
|
|
2021-04-12 07:58:43
|
|
|
2021-04-12 07:58:43
|
|
|
|
|
Deleted User
|
|
andentri
Hey there.
Could someone be kind and explain why I encounter this error :
cjxl -s 9 -m -Q 85 --gaborish=1 -v --override_bitdepth 4 -c 0 -C 0 small.png small.jxl
djxl small.jxl small.png
[...]
`Read 930 compressed bytes [v0.3.7 | SIMD supported: SSE4,Scalar]
Decoded to pixels.
**Failed to write decoded image**.`
is there some inconsistency with the command line options ?
|
|
2021-04-12 08:20:28
|
> `-c 0 -C 0`
What are you doing here? Those options are mutually exclusive...
> `--gaborish=1`
Isn't gaborish reserved for VarDCT?
|
|
|
|
andentri
|
2021-04-12 08:36:47
|
Yes -c was an experiment and should be removed. I don't know about gaborish.
I'm trying to find the best option combination to compress small pictures (40x40).
I'm puzzled why djxl with --bits_per_sample 8 the colors are less good
|
|
2021-04-12 08:36:49
|
|
|
|
andentri
|
|
2021-04-12 08:37:11
|
Look at the elbow
|
|
|
Troc
|
|
BlueSwordM
Honestly, I don't get why people aren't interested in stuff like JPEG-XL.
|
|
2021-04-13 10:37:07
|
"The stuff that's out there works just fine, come to bed honey"
|
|
2021-04-13 10:38:11
|
Still no Android viewer?
|
|
|
|
Deleted User
|
2021-04-13 01:50:11
|
<@&803357352664891472> do you recommend using gaborish at higher (e.g. `-d 1`) quality settings? And what about noise synthesis? I'm trying to make a `-d 1` test encode on 1.19 GiB worth of photos and see how much it'll shrink and if the quality will be indistinguishable from the original.
|
|
|
|
veluca
|
2021-04-13 01:50:46
|
try using default settings 🙂
|
|
2021-04-13 01:51:21
|
I would not use noise for a perceptually-lossless encode target
|
|
|
|
Deleted User
|
2021-04-13 01:52:04
|
So it'd be mainly useful for medium-quality targets (`-d 2` - `-d 4`)?
|
|
|
|
veluca
|
2021-04-13 01:54:59
|
maybe even 4+
|
|
|
|
spider-mario
|
2021-04-13 02:09:55
|
(also, we want to tune it)
|
|
|
|
BlueSwordM
|
2021-04-13 02:29:47
|
Does noise synthesis apply to low frequency and high frequency areas differently?
|
|
2021-04-13 02:39:01
|
IE, is a band filter pass being used to analyze such information and treat it differently?
|
|
|
|
spider-mario
|
2021-04-13 02:51:55
|
synthesis, no (the spectrum is always the same, i.e. blue noise in the loose meaning of the term)
|
|
2021-04-13 02:51:58
|
parameter estimation could
|
|
|
|
Deleted User
|
2021-04-13 02:52:40
|
So the intensity is the only parameter? No noise frequency modulation (like in AV1)?
|
|
|
|
spider-mario
|
2021-04-13 02:53:07
|
right, intensity of noise as a function of brightness, iirc
|
|
2021-04-13 02:53:43
|
the noise is more or less white but high-passed to remove the very lowest frequencies
|
|
|
|
Deleted User
|
2021-04-13 02:54:45
|
It's probably a matter of taste, but just a *little* bit of low-passing that noise would help, too, because in my opinion that noise is a bit too sharp...
|
|
|
spider-mario
right, intensity of noise as a function of brightness, iirc
|
|
2021-04-13 02:58:44
|
That's kinda sad, not every image has the same kind of noise. It'd fail, if e.g. the original noisy image was (for some reason) upscaled before encoding. Film scans are similar in this matter, their grain is not as "sharp" and high-frequency as digital sensor noise that JPEG XL's noise synth is trying to simulate.
|
|
|
|
BlueSwordM
|
2021-04-13 03:00:45
|
To be fair, if high amounts of effort were being put in for that specific task, I'm willing to bet full on noise synthesis could be implemented rather well.
|
|
2021-04-13 03:00:58
|
It could be a goal for a 3rd party JXL encoder... 😛
|
|
|
|
spider-mario
|
2021-04-13 03:01:38
|
I would suspect that low-frequency film grain might be sufficiently adequately represented with the DCT
|
|
2021-04-13 03:02:04
|
with noise synthesis, we focused more on photon shot noise indeed
|
|
|
|
BlueSwordM
|
2021-04-13 03:02:34
|
Ah, so natural noise, correct?
|
|
2021-04-13 03:02:36
|
Interesting.
|
|
|
|
spider-mario
|
2021-04-13 03:02:55
|
yes, the natural fluctuations in photon arrival
|
|
|
|
Deleted User
|
|
BlueSwordM
It could be a goal for a 3rd party JXL encoder... 😛
|
|
2021-04-13 03:03:10
|
I don't think it'd work, JPEG XL's noise synth is probably limited to amplitude only by the standard itself, so I highly doubt if noise synth could be extended to handle frequency modulation...
|
|
|
|
spider-mario
|
2021-04-13 03:03:18
|
most of the visible noise in digital images is in the light itself
|
|
2021-04-13 03:04:14
|
many people think of it as noise that is added by the sensor, but in general that is not the case
|
|
|
|
BlueSwordM
|
|
I don't think it'd work, JPEG XL's noise synth is probably limited to amplitude only by the standard itself, so I highly doubt if noise synth could be extended to handle frequency modulation...
|
|
2021-04-13 03:04:34
|
Well, who said you couldn't just make an alpha layer with all of the noise parameters dialed it with the grain table?
I mean, that's how grain synthesis works in the 1st place: you denoise the image, analyze the difference, and then store the results in a separate grain table layer.
|
|
2021-04-13 03:04:57
|
Theoretically, you could just make an encoder that encodes the noise separately as an extra layer...
|
|
|
|
Deleted User
|
|
BlueSwordM
Well, who said you couldn't just make an alpha layer with all of the noise parameters dialed it with the grain table?
I mean, that's how grain synthesis works in the 1st place: you denoise the image, analyze the difference, and then store the results in a separate grain table layer.
|
|
2021-04-13 03:05:39
|
Oh, that's a good question: how is the noise stored? Aside, as an entirely different set of parameters? Or is it saved in another Modular layer, just like e.g. palette is?
|
|
|
|
spider-mario
|
2021-04-13 03:06:21
|
the noise synthesis currently used by --noise is a set of parameters
|
|
2021-04-13 03:06:30
|
just a look-up table from brightness to noise amplitude
|
|
2021-04-13 03:06:34
|
it takes ~10 bytes iirc
|
|
|
|
Deleted User
|
2021-04-13 03:11:59
|
So with that info provided, I don't think making AV1-like noise synth would succeed in JPEG XL, both technically and practically. Not only we'd have to juggle with additional Modular layers, making older decoders' experience worse, but also it probably wouldn't pay off in terms of quality and efficiency if VarDCT handles lower-frequency noise just fine.
|
|
2021-04-13 03:12:56
|
...does it?
|
|
|
|
spider-mario
|
2021-04-13 03:14:50
|
I must admit that the DCT thing was mostly speculation on my part, I don’t think I have experimented much with such content specifically
|
|
|
|
raysar
|
2021-04-13 04:01:22
|
So to summary the noise synthesis is a separate layer or we can't disable the noise in a jxl with it?
|
|
|
|
Deleted User
|
2021-04-13 04:02:07
|
It's not a layer, it's just a list of parameters for noise synthesizer.
|
|
|
|
raysar
|
2021-04-13 04:03:18
|
Ok, but if i modify a jxl theorically i can disable it?
|
|
|
|
spider-mario
|
2021-04-13 04:03:57
|
you can remove the noise parameters bundle
|
|
2021-04-13 04:04:36
|
I think there is also a decoder option to ignore it
|
|
2021-04-13 04:05:23
|
or maybe not
|
|
|
|
raysar
|
2021-04-13 04:05:26
|
That's cool, we can imagine in a future encoder more noise params like noise frequency etc or it's fixed?
|
|
|
|
spider-mario
|
2021-04-13 04:06:41
|
it’s fixed, but I suppose we could imagine an amendment that would allow another noise model as an extension
|
|
2021-04-13 04:07:01
|
we could then say that that new model would have priority over the current one if both are present
|
|
2021-04-13 04:07:23
|
so old decoders could still synthesize noise the old way
|
|
|
|
_wb_
|
2021-04-13 04:11:43
|
That or we figure out a way to model noise using Modular layers with kMul or kAlphaWeightedAdd
|
|
2021-04-13 04:12:16
|
A lot of noise-like things can be encoded in a short modular image
|
|
|
|
spider-mario
|
|
_wb_
A lot of noise-like things can be encoded in a short modular image
|
|
2021-04-13 04:26:08
|
as we have recently shown in a now dedicated channel
|
|
|
|
Pieter
|
2021-04-13 04:26:12
|
"For some reason, the noise really looks quite a lot like some <#824000991891554375>..."
|
|
|
|
Deleted User
|
2021-04-13 05:09:09
|
<@&803357352664891472> are you sure that `cjxl` frees **all** of the memory it allocated? Even without a bash script I only manage to encode 4-5 files, then the whole WSL gets killed...
|
|
|
|
veluca
|
2021-04-13 05:10:50
|
when a process dies, it releases all of its memory 🙂
|
|
|
|
Deleted User
|
2021-04-13 05:15:13
|
What's even weirder, some of the files are empty!
|
|
2021-04-13 05:18:07
|
For debugging purposes, here's my `jxl_auto.sh`:
```bash
#!/usr/bin/env bash
for file in ~/Zdjęcia/*.JPG; do
~/jpeg-xl/build/tools/cjxl $file ${file%.*}.JXL -j
read -n1 -r -p "Encoded, removing original file soon..."
size=$(stat -c%s "${file%.*}.JXL")
if [ $size -ne 0 ]; then
rm $file
fi
done```
|
|
|
|
spider-mario
|
|
What's even weirder, some of the files are empty!
|
|
2021-04-13 05:36:41
|
could this just be the explorer not refreshing the size?
|
|
|
|
Deleted User
|
2021-04-13 05:37:24
|
Nope, that's after smashing F5 and loading to Chrome Canary (without results).
|
|
|
eustas
|
|
veluca
when a process dies, it releases all of its memory 🙂
|
|
2021-04-13 08:44:55
|
Wasn't that simple in Win95/98/XP times
|
|
|
|
Pieter
|
2021-04-13 08:55:48
|
really?
|
|
|
|
eustas
|
2021-04-13 09:23:41
|
On non-NT windows it was easy to leak "handle" (kind of reference in Java world). Handle had some resources attached to it, e.g. memory. Well, it was fun - there are at most 65535 handles (can't remember for sure, perhaps of each type of handle); one could check all of them to manipulate e.g. windows of other apps (there were no access control as well) 🙂
|
|
|
|
Pieter
|
2021-04-13 09:26:15
|
Oh, I vaguely remember those handles. They're not automatically released when a process terminates?
|
|
|
|
eustas
|
2021-04-13 09:29:40
|
IIRC, no. At least not all of them. "If something is working unusually slow, just reboot" was my recipe then 🙂
|
|
|
|
spider-mario
|
2021-04-13 09:38:17
|
XP was NT 5.1
|
|
2021-04-13 09:38:23
|
Millenium was still DOS-based though
|
|
|
|
Pieter
|
2021-04-13 09:40:08
|
Windows ME was for some reason also the last Windows version I've used as my main OS 😄
|
|
|
|
Deleted User
|
2021-04-13 10:16:40
|
Which OSes are you using now?
|
|
|
|
Pieter
|
2021-04-13 10:17:48
|
Currently Ubuntu; in the past I've used Gentoo and Debian too.
|
|
|
|
BlueSwordM
|
|
Which OSes are you using now?
|
|
2021-04-13 10:19:00
|
Openmandriva
|
|
|
|
Nova Aurora
|
|
Which OSes are you using now?
|
|
2021-04-14 12:42:17
|
Manjaro 🐧
|
|
|
|
veluca
|
|
diskorduser
|
2021-04-14 08:27:05
|
Arch 🐧
|
|
|
|
spider-mario
|
2021-04-14 08:58:17
|
Windows since a few months ago
|
|
2021-04-14 08:58:19
|
Arch before that
|
|
|
|
_wb_
|
2021-04-14 09:23:49
|
Debian
|
|
2021-04-14 09:24:00
|
and MacOS and Android, I guess
|
|
|
OkyDooky
|
2021-04-14 03:01:40
|
Here was your case https://wuvyziz.page.link/uWRfHXjzL3551Drd6
|
|
|
|
spider-mario
|
2021-04-14 03:04:47
|
<@!794205442175402004> we have spam
|
|
|
|
BlueSwordM
|
2021-04-14 03:11:02
|
How did a bot come in 🤔
|
|
|
|
Nova Aurora
|
|
BlueSwordM
How did a bot come in 🤔
|
|
2021-04-14 03:16:17
|
The matrix bridge.
|
|
|
|
_wb_
|
2021-04-14 03:30:43
|
I may have to kill that matrix bridge, if that's how spammers do their thing
|
|
|
|
Nova Aurora
|
|
spider-mario
Windows since a few months ago
|
|
2021-04-14 03:35:39
|
What drove the switch?
|
|
|
|
spider-mario
|
|
Nova Aurora
What drove the switch?
|
|
2021-04-14 04:07:53
|
the hard drive with my `/home` (+ NTFS partition for Windows games) died, and when replacing it and deciding how to split the new space between `/home` and Windows, I thought “I could sidestep the question by using just Windows, and then I wouldn’t have to reboot to change OS, and I could use Secure Boot and full-disk encryption without hassle”
|
|
2021-04-14 04:08:10
|
“and I could play games with correct colors thanks to https://reshade.me/ ”
|
|
|
|
BlueSwordM
|
|
spider-mario
“and I could play games with correct colors thanks to https://reshade.me/ ”
|
|
2021-04-14 04:08:29
|
VKBasalt boi.
|
|
2021-04-14 04:08:41
|
Wait, this is <#794206170445119489> and not <#806898911091753051>
Sorry <:Thonk:805904896879493180>
|
|
|
|
Nova Aurora
|
2021-04-14 04:08:43
|
Gaming, the bane of linux adoption <:PepeHands:808829977608323112>
|
|
|
|
spider-mario
|
2021-04-14 04:09:49
|
well also DxO PhotoLab now
|
|
2021-04-14 04:10:57
|
ah, also I upgraded the GPU while I was at it, and Windows would likely let me use G-Sync more seemlessly
|
|
|
|
Scientia
|
2021-04-14 04:11:11
|
Just paid software in general is the bane of Linux adoption
|
|
2021-04-14 04:11:31
|
There's quite a lot of single platform closed source paid software
|
|
|
|
spider-mario
|
2021-04-14 04:11:33
|
not all of it, e.g. DaVinci Resolve Studio and Reaper have Linux versions
|
|
2021-04-14 04:11:47
|
also pianoteq
|
|
|
|
Scientia
|
2021-04-14 04:11:51
|
That's true, but the same could be said for games
|
|
|
|
Nova Aurora
|
2021-04-14 04:11:54
|
Steam games work well
|
|
|
|
Scientia
|
2021-04-14 04:12:09
|
valve games work well with their native ports
|
|
2021-04-14 04:12:24
|
Proton is a pretty comprehensive solution for gaming
|
|
2021-04-14 04:12:40
|
Technical programs sometimes have issues under wine though
|
|
|
|
Nova Aurora
|
2021-04-14 04:12:40
|
The only reason I play some on windows is modding games on compatibility layers is non-trivial
|
|
|
|
Scientia
|
2021-04-14 04:12:55
|
That's fair
|
|
2021-04-14 04:13:03
|
Oops offtopic lol
|
|
|
|
spider-mario
|
2021-04-14 04:13:27
|
yes, we branched off a bit, sorry 😄
|
|
2021-04-14 04:14:57
|
we can continue in <#806898911091753051> if there is interest
|
|
|
|
Jyrki Alakuijala
|
|
I don't think it'd work, JPEG XL's noise synth is probably limited to amplitude only by the standard itself, so I highly doubt if noise synth could be extended to handle frequency modulation...
|
|
2021-04-14 05:26:28
|
a more complicated noise pattern could be synthesized in another layer using context modeling and/or patches of noise
|
|
|
So with that info provided, I don't think making AV1-like noise synth would succeed in JPEG XL, both technically and practically. Not only we'd have to juggle with additional Modular layers, making older decoders' experience worse, but also it probably wouldn't pay off in terms of quality and efficiency if VarDCT handles lower-frequency noise just fine.
|
|
2021-04-14 05:27:39
|
my understanding of the AV1 noise synth is that it is likely superior in flexibility -- JPEG XL's noise generation is more directed towards nothing will go wrong and it will not spoil/mask information that was sent, and can be made to look very natural
|
|
2021-04-14 05:27:55
|
JPEG XL's noise synth encoding is however badly broken
|
|
2021-04-14 05:28:07
|
it will become better "soon"
|
|
2021-04-14 05:29:09
|
there is a LUT that defines 16 different values of noise, and we have that accidentally reversed in amplitude I believe or we forgot to account for gamma compression in XYB or both
|
|
2021-04-14 05:29:29
|
as a result we have too much noise in white areas and too little in the black areas making things look weird
|
|
|
|
fab
|
2021-04-14 05:35:44
|
try to compress this with jxl
|
|
2021-04-14 05:35:56
|
|
|
2021-04-14 05:36:10
|
from H7m6 playlist
|
|
2021-04-14 05:36:18
|
it has a lot of colours
|
|
2021-04-14 05:36:26
|
this is very difficult for jpeg xl
|
|
2021-04-14 05:36:41
|
without proper encoding decisions it will look subjectively bad
|
|
2021-04-14 05:36:56
|
this is all webp and small resolution
|
|
|
|
Jyrki Alakuijala
|
2021-04-14 05:37:28
|
webp lossy or lossless does well vs jpeg xl?
|
|
|
|
fab
|
2021-04-14 05:37:39
|
this is already compressed and low resolution
|
|
2021-04-14 06:47:56
|
modular -s 8 -I 0.21 -G 2-q 53.19 -X 92.2 -Y 92.19
for %i in (C:\Users\User\Documents\test\*.png) do cjxl4 "%i" "%i.jxl" -s 8 -I 0.21 -q 53.19 -X 92.2 -Y 92.19 -g 2 -m --num_threads=2
|
|
2021-04-14 06:48:48
|
is not loading the jxl
|
|
2021-04-14 06:49:16
|
xnview crashed
|
|
2021-04-14 06:49:25
|
i will re convert
|
|
|
|
_wb_
|
2021-04-14 08:04:19
|
https://c.tenor.com/nwgjBIAl_dAAAAAM/primeval-james-lester.gif
|
|
|
|
fab
|
2021-04-15 02:24:10
|
q 39.3 -P 10 -s 5 -m
|
|
2021-04-15 02:24:13
|
it crashes why?ì
|
|
2021-04-15 02:25:35
|
for %i in (C:\Users\User\Documents\ME3\*.png) do cjxl "%i" "%i.jxl" -q 39.3 -P 10 -s 5 -m --num_threads=2
|
|
2021-04-15 02:26:09
|
|
|
|
|
BlueSwordM
|
|
fab
q 39.3 -P 10 -s 5 -m
|
|
2021-04-15 02:41:17
|
So, can you explain to me why you are trying to do this? <:Thonk:805904896879493180>
|
|
|
|
diskorduser
|
|
BlueSwordM
So, can you explain to me why you are trying to do this? <:Thonk:805904896879493180>
|
|
2021-04-15 04:57:48
|
Blek magic
|
|
|
|
fab
|
2021-04-15 05:17:54
|
https://tipsitaliani.altervista.org/
|
|
|
|
BlueSwordM
|
2021-04-15 06:23:33
|
Fabian, one thing.
|
|
2021-04-15 06:23:49
|
Why not encoding most of your stuff with only one -d XX value instead of trying out multiple?
|
|
|
|
Scientia
|
2021-04-15 07:03:06
|
Why use modular lossy?
|
|
|
|
fab
|
2021-04-15 07:03:07
|
the original is a screenshot
|
|
2021-04-15 07:11:48
|
91,6 kb all two images
|
|
|
|
Scientia
|
2021-04-15 07:20:16
|
has anyone done tests on decompression speeds with different speed values?
|
|
2021-04-15 07:20:53
|
like if you need the smallest size, is there a particularly bad penalty on decompression for using -s 9
|
|
|
|
fab
|
2021-04-15 07:27:02
|
something like that should be better
|
|
2021-04-15 07:27:03
|
-d 3.39 -gaborish=1 -dots=1 --epf=2 noise=0 -s 6
|
|
2021-04-15 07:27:08
|
simpler is better
|
|
|
|
veluca
|
2021-04-15 07:31:43
|
faster decoding does nothing in modular mode
|
|
2021-04-15 07:32:14
|
-s 9 shouldn't be *significantly* slower but I don't really remember numbers
|
|
|
|
fab
|
2021-04-15 07:33:55
|
in fact it was 1,6 mpx/s
|
|