|
fab
|
2023-06-07 07:05:38
|
Or better I 0.174 photon noise 552 d 0.466 e 8
|
|
2023-06-07 07:38:37
|
The bug with jxl is that quality is worth it at e7 d 0.714 but only if you use progressive
|
|
|
yoochan
|
2023-06-07 07:39:27
|
what ?
|
|
|
|
fab
|
2023-06-07 07:39:35
|
Av2 looks terrible I have tried p9 of compound prediction not inpressed at lot
|
|
2023-06-07 07:41:10
|
Q 59.32 s8 photon noise 175 i 0.178 patches 0
|
|
|
|
yoochan
|
2023-06-07 07:41:24
|
jpegxl killer feature for me is: lossless conversion from jpeg (and png, ang gif) the only slightly better quality than av1 is just sugar coating 🙂
|
|
|
|
fab
|
2023-06-07 07:41:40
|
Would yield to a similar images to the one of av2
|
|
2023-06-07 07:44:20
|
d 0.599 -I 0.941 e 7 photon noise 949
|
|
2023-06-07 07:50:00
|
40.389
|
|
2023-06-07 07:50:05
|
52.8
|
|
2023-06-07 07:53:23
|
To me av2 at cpu 0 Min 13 max 39
|
|
2023-06-07 07:53:35
|
It has already 9,2% gains
|
|
2023-06-07 07:57:43
|
At d 0.544 s7. q 91.44 s9 p 83.14 a 465 I 0.441 target intensity 237 gaborish 0
|
|
2023-06-07 07:58:36
|
Qp 78 max 127
|
|
2023-06-07 07:59:20
|
7,81% gains to jxl
|
|
2023-06-07 08:03:06
|
26,37% to s9 d 0.909 I 0.57 photon noise 664
|
|
2023-06-07 08:04:17
|
In screenshots condition even higher
|
|
2023-06-07 08:06:18
|
Comparing to video 36.58% svt p8 15.2crf
|
|
2023-06-07 08:06:47
|
22,744% to AV1
|
|
2023-06-07 08:07:10
|
41,1% to vp9
|
|
2023-06-07 08:07:47
|
68,27% to x264 slower
|
|
2023-06-07 08:20:23
|
cjxl -e 9 -q 83.14 --photon_noise_iso=465 -I 0.441 --intensity_target=237 --gaborish=0 --lossless_jpeg=0 C:\Users\Use\Documents\dn\ubde.png C:\Users\Use\Documents\dn\ubde.jxl
cjxl -e 9 -q 83.14 --photon_noise_iso=465 -I 0.441 --intensity_target=237 --gaborish=0 --lossless_jpeg=0 C:\Users\Use\Documents\jxl\a.jpg C:\Users\Use\Documents\jxl\weea.jxl
|
|
|
DZgas Ж
|
2023-06-07 09:17:43
|
<:PepeGlasses:878298516965982308>
|
|
|
Quackdoc
|
|
Quackdoc
just tried it again, it's pretty close but the decode speed penalty is pretty bad still the size makes it quite worth it though
|
|
2023-06-07 09:29:25
|
well seems like i'm hitting a bug with ffmpeg when trying to convert videos into a image sequences muxed in mkv or nut, so I can't compare quality since im too lazy to make an img sequence then mux that like that
|
|
|
|
Sakaria
|
|
_wb_
oh, interesting. How many components do you have originally? It could be interesting to see what happens if you don't reduce the number of components, jxl can represent up to 2^12 channels...
|
|
2023-06-07 09:48:25
|
the images have 120 components originally. Couldn't find an easy way to compress images with more than 4 components using the tools. Is it possible?
|
|
|
_wb_
|
2023-06-07 10:41:03
|
Not with cjxl, but with libjxl you can add 117 extra channels besides rgb (or 119 + gray)
|
|
|
|
Sakaria
|
2023-06-07 12:13:19
|
For those who are interested this is the output when running cjxl input.png out.jxl --target_bpp=4 -e 7
|
|
2023-06-07 12:14:57
|
When i try to run it lossless with cjxl input.png out.jxl -q 100 -e 7 . The output is this
|
|
2023-06-07 12:17:21
|
It seems like the encoder can't achieve bpps between 4 and 0.6 which is unexpected at least to me since the MSE is quite large even with distance 0.010
|
|
|
|
_wb_
|
2023-06-07 12:55:47
|
it doesn't go below d0.01 which should be _visually_ very high quality — but your data is not visual...
|
|
2023-06-07 01:04:15
|
you could do something like `--intensity_target=100000` to make it produce higher quality lossy, but there's still the problem that it will not be optimizing for MSE but for perceptual quality when interpreting your data as sRGB colors
|
|
|
jonnyawsom3
|
2023-06-07 01:12:14
|
I'm just curious, what's the aim of this? To produce high quality but still lossy files? Because it looks like lossless already reached your 4 bpp target and without loosing any data
|
|
|
|
_wb_
|
2023-06-07 01:23:31
|
he wants to go below 4 bpp — and the highest quality lossy is 0.6 bpp
|
|
2023-06-07 01:24:45
|
I suppose this is a filmstrip image consisting of 120 sub-images that are 684x956, each representing one component of a 120-channel image
|
|
2023-06-07 01:29:38
|
what's the range for the values you have in input.png, <@687984546553266176> ? Are you perhaps using a 16-bit png file but the actual range of the values is basically around 0 (or maybe around 2^15, if you're using signed values with an offset to fit it in png?)
|
|
|
|
Sakaria
|
2023-06-07 01:31:59
|
Yeah, you are correct, most of the values will be around 2^15
|
|
|
|
_wb_
|
2023-06-07 01:35:28
|
perhaps you can rescale it to use more of the range? You could also consider using pfm input so you can use actual signed values
|
|
2023-06-07 01:39:01
|
it now probably just looks like an image that is mostly 50% gray or very close to it, and perceptual encoding of that will be not really what you want
|
|
|
|
yoochan
|
|
Sakaria
Yeah, you are correct, most of the values will be around 2^15
|
|
2023-06-07 01:55:34
|
I would be curious to see an exemple of a channel... of an histogram of it if it's too secretive
|
|
|
|
jonnyawsom3
|
2023-06-07 01:58:26
|
Might not matter, but do bear in mind they're on version 0.6 to use the bpp target
|
|
2023-06-07 01:59:40
|
Actually, would modular lossy have any different effect?
|
|
|
|
_wb_
|
2023-06-07 02:11:10
|
lossy modular also uses XYB so that's not ideal, but perhaps using pfm input with `-x color_space=XYB_Per` (so it treats the image as already being in XYB space) could work...
|
|
|
Jyrki Alakuijala
|
2023-06-07 02:52:53
|
https://github.com/libjxl/libjxl/pull/2537 gives another 0.13 % for JPEG XL vardct quality 😛
|
|
2023-06-07 02:53:09
|
small steps
|
|
|
|
yoochan
|
2023-06-07 02:54:22
|
how large is your corpus ? don't you fear to over-optimize for it ?
|
|
|
|
fab
|
2023-06-07 04:57:22
|
|
|
2023-06-07 04:57:49
|
Compress this with jxl
|
|
2023-06-07 04:58:00
|
Latest build by jyrki
|
|
2023-06-07 04:58:24
|
Send jxl file and parameters
|
|
2023-06-07 05:01:50
|
I'll say 77,387 e7 is worth it
|
|
|
Fox Wizard
|
2023-06-07 05:01:53
|
Blur™️
|
|
|
|
fab
|
2023-06-07 05:02:16
|
It hides my defects at taking photos
|
|
2023-06-07 05:02:49
|
But I want to know Can jxl detect if a photos was shoot by an autistic
|
|
|
|
Fox Wizard
|
2023-06-07 05:03:34
|
<:Kekw:1098190857267576923>
|
|
|
|
fab
|
2023-06-07 05:04:35
|
Or e8 d 0.723 for normal screenshots
|
|
2023-06-07 05:09:37
|
|
|
2023-06-07 05:09:59
|
-8 titling technique I Learned
|
|
2023-06-07 05:12:37
|
S 8 0.608 0.193 target intensity 254
|
|
2023-06-07 05:12:57
|
With a newer build than that
|
|
2023-06-07 05:13:01
|
804047396522426378
|
|
2023-06-07 05:14:10
|
0.426 e8 the quality is incredible
|
|
|
Traneptora
|
|
Quackdoc
I tried this a while ago by muxing a jxl sequences into mkv, it was quite the while ago, but it worked really well aside from decode speed, but libjxl was pretty immature then
|
|
2023-06-07 05:15:59
|
it's theoretical tbh, not something you can do right now as a drop-in-replacement
|
|
|
Sakaria
For those who are interested this is the output when running cjxl input.png out.jxl --target_bpp=4 -e 7
|
|
2023-06-07 05:17:08
|
that's a very old version, 0.6.1
|
|
2023-06-07 05:17:17
|
try updating to a recent release, or git master
|
|
|
|
fab
|
2023-06-07 05:17:35
|
Jxl in this way get a Yellow tint
|
|
2023-06-07 05:17:38
|
Like urine
|
|
2023-06-07 05:21:54
|
D q 93.24 e 9 I 0.561 target intensity 223 photon noise 669 epf 0
|
|
|
|
Quackdoc
|
|
Traneptora
it's theoretical tbh, not something you can do right now as a drop-in-replacement
|
|
2023-06-07 05:26:20
|
sorry for brevity, on phone can add jxl to riff bmp tags to get ffmpeg to mux jxl into mkv, mov mp4 etc.
the decode speed is a little low for my system (400fps in mpv) sounds like a lot but in reality need around 600fps for really good scrubbing still usable, don't get me wrong.
biggest issue is that ffmpeg seems to be messing up some conversions, so some videos get really mangled (happens with png and even raw video, no idea why)
```patch
diff --git a/libavformat/riff.c b/libavformat/riff.c
index df7e9df31b..16e37fb557 100644
--- a/libavformat/riff.c
+++ b/libavformat/riff.c
@@ -34,6 +34,7 @@
* files use it as well.
*/
const AVCodecTag ff_codec_bmp_tags[] = {
+ { AV_CODEC_ID_JPEGXL, MKTAG('J', 'X', 'L', ' ') },
{ AV_CODEC_ID_H264, MKTAG('H', '2', '6', '4') },
{ AV_CODEC_ID_H264, MKTAG('h', '2', '6', '4') },
{ AV_CODEC_ID_H264, MKTAG('X', '2', '6', '4') },
```
|
|
2023-06-07 05:27:29
|
works best in nut for some reason some video editors like olive don't work when its muxed in mkv but do in nut. I plan on investigating
|
|
|
|
Traneptora
|
2023-06-07 05:27:57
|
many editors have their own matroska demuxers
|
|
2023-06-07 05:28:01
|
but rely on avformat for nut
|
|
|
Quackdoc
sorry for brevity, on phone can add jxl to riff bmp tags to get ffmpeg to mux jxl into mkv, mov mp4 etc.
the decode speed is a little low for my system (400fps in mpv) sounds like a lot but in reality need around 600fps for really good scrubbing still usable, don't get me wrong.
biggest issue is that ffmpeg seems to be messing up some conversions, so some videos get really mangled (happens with png and even raw video, no idea why)
```patch
diff --git a/libavformat/riff.c b/libavformat/riff.c
index df7e9df31b..16e37fb557 100644
--- a/libavformat/riff.c
+++ b/libavformat/riff.c
@@ -34,6 +34,7 @@
* files use it as well.
*/
const AVCodecTag ff_codec_bmp_tags[] = {
+ { AV_CODEC_ID_JPEGXL, MKTAG('J', 'X', 'L', ' ') },
{ AV_CODEC_ID_H264, MKTAG('H', '2', '6', '4') },
{ AV_CODEC_ID_H264, MKTAG('h', '2', '6', '4') },
{ AV_CODEC_ID_H264, MKTAG('X', '2', '6', '4') },
```
|
|
2023-06-07 05:28:41
|
which conversions, colorspace?
|
|
|
|
Quackdoc
|
2023-06-07 05:28:45
|
in olive we use ffmpeg for demuxing most things, mkv included I think, may be mistaken
|
|
|
|
Traneptora
|
2023-06-07 05:29:16
|
swscale is not known for a high degree of accuracy when doing that
|
|
|
|
Quackdoc
|
|
Traneptora
which conversions, colorspace?
|
|
2023-06-07 05:29:28
|
I have no idea, at first I thought so, but after some testing I think the decoding of vp9 itself is the issue, also tried libplacebo to no avail
|
|
2023-06-07 05:29:55
|
exact same issue when using ffmpegs diff to nearly a T
|
|
|
|
Traneptora
|
2023-06-07 05:29:56
|
placebo does colorspace conversions very well
|
|
|
|
Quackdoc
|
2023-06-07 05:30:09
|
yeah, is why I think decoding is bugged
|
|
|
|
Traneptora
|
2023-06-07 05:30:09
|
what command with ffmpeg messes it up?
|
|
2023-06-07 05:30:32
|
did you try forcing the libvpx-vp9 decoder instead of the ffvp9 internal decoder?
|
|
|
|
Quackdoc
|
2023-06-07 05:30:45
|
I didn't, ill certainly try that
|
|
2023-06-07 05:31:07
|
tbh I kinda forgot you can do that
|
|
|
|
Traneptora
|
|
Quackdoc
sorry for brevity, on phone can add jxl to riff bmp tags to get ffmpeg to mux jxl into mkv, mov mp4 etc.
the decode speed is a little low for my system (400fps in mpv) sounds like a lot but in reality need around 600fps for really good scrubbing still usable, don't get me wrong.
biggest issue is that ffmpeg seems to be messing up some conversions, so some videos get really mangled (happens with png and even raw video, no idea why)
```patch
diff --git a/libavformat/riff.c b/libavformat/riff.c
index df7e9df31b..16e37fb557 100644
--- a/libavformat/riff.c
+++ b/libavformat/riff.c
@@ -34,6 +34,7 @@
* files use it as well.
*/
const AVCodecTag ff_codec_bmp_tags[] = {
+ { AV_CODEC_ID_JPEGXL, MKTAG('J', 'X', 'L', ' ') },
{ AV_CODEC_ID_H264, MKTAG('H', '2', '6', '4') },
{ AV_CODEC_ID_H264, MKTAG('h', '2', '6', '4') },
{ AV_CODEC_ID_H264, MKTAG('X', '2', '6', '4') },
```
|
|
2023-06-07 05:32:37
|
btw I just merged a patch from the ffmpeg ML last week that changes how the libjxl decoder wrapper works, it might be faster. try enchmarking against git master FFmpeg too
|
|
|
|
Quackdoc
|
|
Traneptora
btw I just merged a patch from the ffmpeg ML last week that changes how the libjxl decoder wrapper works, it might be faster. try enchmarking against git master FFmpeg too
|
|
2023-06-07 05:38:26
|
will do, the command I use to replicate the issue is as simple as `ffmpeg -i unison.mkv -c:v png /nvme/junk/test.mkv` and `ffmpeg -i unison.mkv -c:v rawvideo /nvme/junk/test.nut ` the later should prevent colorspace conversion from happening as far as i know.
as for comparison i was testing ssimulacra2rs, but for a quick example I just use
`mpv unison.mkv --external-file=/nvme/junk/test.nut --profile=diff`
where profile=diff /diffb is
```toml
[diff]
lavfi-complex='[vid1] [vid2] blend=all_mode=difference [vo]'
[diffb] #increase gamma to help visibility for low diff images
lavfi-complex='[vid1] [vid2] blend=all_mode=difference[d], [d] eq=gamma=2[vo]'
```
|
|
|
|
Traneptora
|
|
Quackdoc
will do, the command I use to replicate the issue is as simple as `ffmpeg -i unison.mkv -c:v png /nvme/junk/test.mkv` and `ffmpeg -i unison.mkv -c:v rawvideo /nvme/junk/test.nut ` the later should prevent colorspace conversion from happening as far as i know.
as for comparison i was testing ssimulacra2rs, but for a quick example I just use
`mpv unison.mkv --external-file=/nvme/junk/test.nut --profile=diff`
where profile=diff /diffb is
```toml
[diff]
lavfi-complex='[vid1] [vid2] blend=all_mode=difference [vo]'
[diffb] #increase gamma to help visibility for low diff images
lavfi-complex='[vid1] [vid2] blend=all_mode=difference[d], [d] eq=gamma=2[vo]'
```
|
|
2023-06-07 05:39:26
|
makes sense
|
|
|
|
Quackdoc
|
2023-06-07 05:41:30
|
hmm. just tested avc video, seems to be presenting the exact same issue maybe ill recompile rq
|
|
|
|
Traneptora
|
2023-06-07 05:42:52
|
`[vid1] [vid2] blend=all_mode=difference[vo]` followed by `--gamma=100` as another option
|
|
2023-06-07 05:42:59
|
also works btw
|
|
2023-06-07 05:43:10
|
that does the gamma correction on the GPU rather than in an avfilter filter
|
|
|
Quackdoc
will do, the command I use to replicate the issue is as simple as `ffmpeg -i unison.mkv -c:v png /nvme/junk/test.mkv` and `ffmpeg -i unison.mkv -c:v rawvideo /nvme/junk/test.nut ` the later should prevent colorspace conversion from happening as far as i know.
as for comparison i was testing ssimulacra2rs, but for a quick example I just use
`mpv unison.mkv --external-file=/nvme/junk/test.nut --profile=diff`
where profile=diff /diffb is
```toml
[diff]
lavfi-complex='[vid1] [vid2] blend=all_mode=difference [vo]'
[diffb] #increase gamma to help visibility for low diff images
lavfi-complex='[vid1] [vid2] blend=all_mode=difference[d], [d] eq=gamma=2[vo]'
```
|
|
2023-06-07 05:44:26
|
did you pick the right decoder when doing the comparison?
|
|
2023-06-07 05:44:32
|
and/or are you hardware decoding?
|
|
|
|
Quackdoc
|
2023-06-07 05:48:14
|
software dec, I just confirmed it's happening with avc input videos as well, and doesn't seem to be present when encoding with libx264
|
|
2023-06-07 05:49:09
|
just recompiled and didn't seem to fix the issue either
|
|
|
|
Traneptora
|
2023-06-07 05:50:14
|
this sounds like a log file is in order
|
|
2023-06-07 05:50:33
|
also should probably head to the #ffmpeg irc channel
|
|
2023-06-07 05:50:39
|
as <#794206170445119489> is not really the place for this
|
|
|
|
Quackdoc
|
2023-06-07 05:50:50
|
yeah ill do so
|
|
|
|
afed
|
|
Jyrki Alakuijala
https://github.com/libjxl/libjxl/pull/2537 gives another 0.13 % for JPEG XL vardct quality 😛
|
|
2023-06-07 06:10:58
|
also, this is still not merged
https://github.com/libjxl/libjxl/pull/2348
|
|
|
|
fab
|
2023-06-08 06:54:29
|
Aomenc 3.6.1-712
|
|
2023-06-08 06:54:36
|
Has no blurring
|
|
2023-06-08 06:54:49
|
At CPU 8
|
|
2023-06-08 06:54:57
|
Qp 26 10bit
|
|
2023-06-08 06:55:26
|
I will try also av2
|
|
|
|
DZgas Ж
|
2023-06-08 06:55:30
|
<:Thonk:805904896879493180> <#794206170445119489>
|
|
|
|
fab
|
2023-06-08 07:02:08
|
The quality of video is poor
|
|
|
|
Jyrki Alakuijala
|
|
afed
also, this is still not merged
https://github.com/libjxl/libjxl/pull/2348
|
|
2023-06-08 10:34:28
|
I'll try reworking this, looks like I forgot about this
|
|
|
|
Quackdoc
|
2023-06-08 11:09:11
|
finally got it, using jxl as an alternative to prores in olive, caching is disabled quality is d0.5 which I find to be roughly the same as prores_ks hq e3
```
2.9G quack 8 Jun 06:23 chim-jxk.mkv
7.3G quack 8 Jun 06:09 chim-pro.mov
```
https://cdn.discordapp.com/attachments/587033245061873759/1116322453786984458/olive-jxl.mp4
|
|
2023-06-08 11:10:21
|
scrubbing perf is... adequate, not nearly as good as prores, but meh, it is what it is, running libjxl `8938239` and ffmpeg `141d11c` since the commit that added animated jxl decoding broke decoding for some reason with the patch I use, not sure which patch, I tested the commit's just before it and just after it
|
|
2023-06-08 11:13:09
|
speaking of the patch, is there any reason (aside from the animated jxl breaking it) not to add riff tags for jxl in ffmpeg?
|
|
|
|
Traneptora
|
|
Quackdoc
scrubbing perf is... adequate, not nearly as good as prores, but meh, it is what it is, running libjxl `8938239` and ffmpeg `141d11c` since the commit that added animated jxl decoding broke decoding for some reason with the patch I use, not sure which patch, I tested the commit's just before it and just after it
|
|
2023-06-08 11:14:43
|
how is it "broken" exactly? I switched over from the decode_frame internal method to the receive_frame internal method
|
|
|
|
Quackdoc
|
2023-06-08 11:15:00
|
it can only decode a single frame then hangs
|
|
|
|
Traneptora
|
2023-06-08 11:15:13
|
are you feeding it more than one AVPacket?
|
|
|
|
Quackdoc
|
2023-06-08 11:15:53
|
im not sure how the internals of it work, I just added JXL tags to riff bmp tags so I can mux and demux it like that,
|
|
|
Quackdoc
sorry for brevity, on phone can add jxl to riff bmp tags to get ffmpeg to mux jxl into mkv, mov mp4 etc.
the decode speed is a little low for my system (400fps in mpv) sounds like a lot but in reality need around 600fps for really good scrubbing still usable, don't get me wrong.
biggest issue is that ffmpeg seems to be messing up some conversions, so some videos get really mangled (happens with png and even raw video, no idea why)
```patch
diff --git a/libavformat/riff.c b/libavformat/riff.c
index df7e9df31b..16e37fb557 100644
--- a/libavformat/riff.c
+++ b/libavformat/riff.c
@@ -34,6 +34,7 @@
* files use it as well.
*/
const AVCodecTag ff_codec_bmp_tags[] = {
+ { AV_CODEC_ID_JPEGXL, MKTAG('J', 'X', 'L', ' ') },
{ AV_CODEC_ID_H264, MKTAG('H', '2', '6', '4') },
{ AV_CODEC_ID_H264, MKTAG('h', '2', '6', '4') },
{ AV_CODEC_ID_H264, MKTAG('X', '2', '6', '4') },
```
|
|
2023-06-08 11:16:07
|
this patch here
|
|
|
|
Traneptora
|
2023-06-08 11:16:07
|
each frame needs to be its own AVPacket
|
|
|
|
fab
|
2023-06-08 11:16:20
|
Use q95 e9
|
|
|
|
Traneptora
|
2023-06-08 11:16:23
|
this wasn't true before my patch but it *should* still be true
|
|
|
|
fab
|
2023-06-08 11:16:30
|
Not d 0.5
|
|
|
|
Traneptora
|
2023-06-08 11:16:31
|
i.e. you should still be doing that
|
|
2023-06-08 11:16:56
|
it's looking like randomly adding jxl to riff tags doesn't create a separate AVPacket for each frame
|
|
|
|
fab
|
2023-06-08 11:16:59
|
Also use the commit by jon Sneyers fix low precision text
|
|
|
|
Traneptora
|
2023-06-08 11:17:00
|
that's the issue
|
|
|
|
fab
|
2023-06-08 11:17:11
|
Will Speed up encoding also
|
|
|
|
Traneptora
|
|
Quackdoc
speaking of the patch, is there any reason (aside from the animated jxl breaking it) not to add riff tags for jxl in ffmpeg?
|
|
2023-06-08 11:18:08
|
because ideally you have one frame be one packet, which doesn't appear to be happening when you just mark it as supported and don't add any extra casework
|
|
|
|
Quackdoc
|
2023-06-08 11:18:22
|
well, I never expected it to be stable at least lol
|
|
2023-06-08 11:18:37
|
it was more or less a quick hack for toying around
|
|
|
spider-mario
|
2023-06-08 12:05:18
|
is it intended as a format for the actual source or for proxies?
|
|
|
|
Quackdoc
|
2023-06-08 12:08:34
|
I'm treating it as a mezzanine, so encode the source into it then work with the jxl video directly instead of a proxy or the source, though it would likely be suitable for a proxy as well, it's not something I have yet tested
|
|
|
|
jonnyawsom3
|
|
fab
Use q95 e9
|
|
2023-06-08 02:39:35
|
That's worse quality than they had set, and I think they wanted the video to be made within this week ideally
|
|
|
|
Quackdoc
|
2023-06-08 02:43:42
|
I tried q95 e4, got a bit smaller and a bit better then before, but still not quite to the level of prores. linked is prores on top, jxl on bottom using scores from ssimulacra2rs, not sure how much it matters but I am using netflix's chimera av1 encode, so it's going from yuv420p10le -> rgb480 -> yuv444p10le due to a big in ssimulacra2 not working with anything but yuv input
https://cdn.discordapp.com/attachments/587033245061873759/1116353467326410832/prores.png
https://cdn.discordapp.com/attachments/587033245061873759/1116374917995315290/jxlseqQ95E4.png
|
|
2023-06-08 02:44:32
|
it took about 2x the encode time lol at the very least its `2.70 GiB chim-jxk.mkv` vs `6.77 GiB chim-pro.mov` but going e5+ is simply too slow
|
|
2023-06-08 02:44:46
|
tho, I could try av1an I suppose
|
|
|
Oleksii Matiash
|
2023-06-08 03:05:56
|
Found quite non-obvious (for me) fact: lossy compression from 16 bpp source produces file with better quality (not visible without difference amplifiaction though) AND smaller size, than the same 16 bpp source converted to 8 bpp by photoshop and compressed with the same settings
|
|
|
|
fab
|
|
That's worse quality than they had set, and I think they wanted the video to be made within this week ideally
|
|
2023-06-08 03:23:43
|
Then use e8 d 0.504 and Jon precision branch for screenshots of a phone
|
|
2023-06-08 03:24:11
|
But anyway the compression isn't higher than webp2 at q90
|
|
2023-06-08 03:24:29
|
And i didn't expected that
|
|
|
|
_wb_
|
|
Oleksii Matiash
Found quite non-obvious (for me) fact: lossy compression from 16 bpp source produces file with better quality (not visible without difference amplifiaction though) AND smaller size, than the same 16 bpp source converted to 8 bpp by photoshop and compressed with the same settings
|
|
2023-06-08 03:27:50
|
it's true for basically the same reason as what happens when you're converting to GIF (or PNG8) first and then doing lossy compression instead of starting from PNG24.
|
|
|
|
Oleksii Matiash
|
|
_wb_
it's true for basically the same reason as what happens when you're converting to GIF (or PNG8) first and then doing lossy compression instead of starting from PNG24.
|
|
2023-06-08 03:38:18
|
Yes, now it is obvious for me 🙂 I just used to the think "lossy = 8 bpp" because of jpeg, as the conversion 16 -> 8 bpp was one of the steps in my "original 16 bpp tiff -> small sRGB jpeg" flow. Maybe it is worth adding to the docs somewhere? Or it is already somewhere and I just missed it?
|
|
|
|
_wb_
|
2023-06-08 03:51:09
|
I don't think it's written anywhere (and I wouldn't really know what would be the best place for that). But it's likely something that will solve itself as more and more tooling gets direct jxl support so intermediate files can be avoided.
|
|
|
|
jonnyawsom3
|
2023-06-08 03:57:03
|
It'll probably just take time to get the old stigmas of the original jpeg out people's heads, like lossy only, 8 bit, SDR, ect
|
|
|
|
_wb_
|
2023-06-08 03:59:43
|
sure — that's one of the disadvantages of having "JPEG" in the name, people associate it with the de facto JPEG 1 standard and its implementations, which are lossy only, 8-bit only, etc
|
|
|
|
spider-mario
|
2023-06-08 04:14:39
|
I always keep everything 16-bit until the very final compression step, I don’t see the point of quantizing earlier on
|
|
|
|
Traneptora
|
2023-06-08 04:40:42
|
GIMP 210 can work in linear light, which is nice
|
|
2023-06-08 04:41:02
|
unfortunately it's awkward if you want to export sRGB 8bit PNG
|
|
2023-06-08 04:41:18
|
since it exports 8bit linear with a linear icc
|
|
2023-06-08 04:41:48
|
can always use placebo tho
|
|
|
elfeïn
|
|
_wb_
sure — that's one of the disadvantages of having "JPEG" in the name, people associate it with the de facto JPEG 1 standard and its implementations, which are lossy only, 8-bit only, etc
|
|
2023-06-08 05:18:56
|
shoulda stuck with fliff
|
|
|
|
Traneptora
|
2023-06-08 05:36:06
|
why?
|
|
2023-06-08 05:36:12
|
JXL isn't a "free lossless image format"
|
|
|
|
elfeïn
|
2023-06-08 05:47:49
|
What does jpeg mean?
|
|
2023-06-08 05:47:53
|
or xl?
|
|
|
|
Traneptora
|
|
elfeïn
What does jpeg mean?
|
|
2023-06-08 05:51:11
|
Joint Photographic Experts Group
|
|
2023-06-08 05:51:30
|
a lot of the newer codecs created by the JPEG committee have "X" in the name
|
|
2023-06-08 05:51:32
|
so that's what X is for
|
|
2023-06-08 05:51:34
|
L is for long-term
|
|
2023-06-08 05:51:45
|
I'd like to say it stands for eXtended-Life but it doesn't
|
|
|
|
jonnyawsom3
|
2023-06-08 05:54:57
|
Xtra Longterm
|
|
|
|
_wb_
|
2023-06-08 05:57:42
|
excellent
|
|
|
|
elfeïn
|
2023-06-08 06:02:00
|
well there you go
|
|
2023-06-08 06:02:09
|
nobody cares about the acronym
|
|
2023-06-08 06:03:01
|
you could call it the ultimate image format, uif
|
|
2023-06-08 06:03:06
|
wiff
|
|
2023-06-08 06:03:36
|
~~or Yet-another Image Format~~
|
|
|
|
DZgas Ж
|
2023-06-08 06:08:18
|
Juggernauts' Perfectly Enduring Graphic Xperience of Lifelong enjoyment
Joyous and Profoundly Eternal Graphics eXtended Longevity format
Just Perpetuate Endless Greatness with eXtreme Longevity
Jaw-droppingly Perfect and Everlasting Graphics in eXtra Large format
Jubilant Preservation of Everlasting Greatness through eXpanded Lifespan.
|
|
2023-06-08 06:08:37
|
👉 neuralgen
|
|
2023-06-08 06:09:10
|
eXtra Large <:JXL:805850130203934781>
|
|
|
|
spider-mario
|
2023-06-08 06:39:58
|
**j**e **p**eux **e**nfin **g**oûter à l’e**x**tra **l**uxueux
|
|
|
|
fab
|
2023-06-08 07:23:19
|
|
|
2023-06-08 07:23:21
|
help fixing that
|
|
2023-06-08 07:53:15
|
|
|
2023-06-08 07:53:28
|
My xiaomi phone is burning anymore
|
|
2023-06-08 07:53:58
|
is it not taking any good photo anymore
|
|
2023-06-08 07:54:52
|
My mother sometimes one day a week before i go to bed shake the phone because i stay too much
|
|
2023-06-08 07:55:11
|
And at 22:15 she considers it late
|
|
2023-06-08 07:55:45
|
Because my father is old and he's going to countryside
|
|
2023-06-08 07:56:13
|
He bring lemon for 0.16 at 3,20 for 20 kilos
|
|
2023-06-08 07:56:28
|
The economy of Italy isn't bad
|
|
2023-06-08 07:56:47
|
The problem is young people don't wanna do anything anymore
|
|
2023-06-08 07:57:01
|
28% don't work
|
|
2023-06-08 07:57:30
|
And 14,5% of the girls didn't want to find a job
|
|
2023-06-08 07:57:33
|
A job
|
|
2023-06-08 07:57:35
|
Not a work
|
|
2023-06-08 07:57:49
|
I know the words
|
|
2023-06-08 07:58:20
|
The problem is there isn't any discussion setted
|
|
2023-06-08 07:58:24
|
Is all autism
|
|
|
lonjil
|
2023-06-08 08:30:52
|
Personally I've been trying to use "JXL" as the only name I use for the format.
|
|
2023-06-08 08:31:10
|
Because it looks cool and doesn't spell out "JPEG"
|
|
|
|
veluca
|
|
lonjil
Personally I've been trying to use "JXL" as the only name I use for the format.
|
|
2023-06-08 09:04:07
|
I use that a lot too
|
|
|
VcSaJen
|
2023-06-09 02:11:44
|
I use "JPEG XL (JXL)"
|
|
|
username
|
2023-06-09 02:15:15
|
I have noticed that as time has went on I have been using "JXL" way more then "JPEG XL"/"JPEG-XL" probably because it's a lot less effort to type
|
|
|
|
elfeïn
|
2023-06-09 02:21:48
|
I thought it was called JXL this whole time.
|
|
|
OkyDooky
|
2023-06-09 03:36:13
|
<@794205442175402004>\: On the WebKitGTK (i.e. GNOME and embedded) side\: *It's happening!* https://bugs.webkit.org/show_bug.cgi?id=257871
|
|
|
|
jonnyawsom3
|
2023-06-09 04:05:04
|
JXL is both shorter and less connected to old Jpeg. I can't count the amount of times I've searched for "JpegXL" or "Jpeg XL" just to get a "Did you mean jpeg?" or just showing the wrong results anyway
|
|
|
|
Jyrki Alakuijala
|
2023-06-09 09:44:18
|
Let jixel store all your pixel.
|
|
2023-06-09 09:45:30
|
in 20 years some people will discuss if JXL should be pronounced jixel or gixel 😄
|
|
|
|
_wb_
|
2023-06-09 09:46:26
|
it should of course be pronounced yixel
|
|
|
|
Jyrki Alakuijala
|
2023-06-09 09:51:46
|
I'd go for dʒɪksəl as a pronunciation guide
|
|
|
|
_wb_
|
2023-06-09 09:54:53
|
if the J is for JPEG / Joint then that makes sense
|
|
2023-06-09 09:55:44
|
if the J is for Jyrki / Jon / Jan however... 🙂
|
|
|
|
yoochan
|
2023-06-09 09:59:45
|
djéxel then like djépeg
|
|
|
|
_wb_
|
2023-06-09 10:01:10
|
djéxèlle
|
|
|
|
fab
|
2023-06-09 10:05:01
|
|
|
2023-06-09 10:06:00
|
Ji as the author that is making the codec
|
|
2023-06-09 10:06:16
|
Not the one that optimise faster speeds
|
|
2023-06-09 10:06:26
|
Or precision alone
|
|
2023-06-09 10:06:36
|
I trust more jyrki
|
|
2023-06-09 10:07:35
|
It doesn't change anything in the meaning
|
|
|
|
Jyrki Alakuijala
|
2023-06-09 10:26:28
|
J for Jan/Jon/Jyrki, X for Zoltan and Moritz and L for Lode and Luca (-:
|
|
2023-06-09 10:27:09
|
pronunciation guide for JXL is: 'jajojyzomololu'
|
|
|
|
Traneptora
|
2023-06-09 10:45:34
|
I'm gonna start calling it jajojyzomololu
|
|
2023-06-09 10:45:53
|
~~ok no im not~~
|
|
|
|
Jyrki Alakuijala
|
2023-06-09 01:01:11
|
ops, I forgot spidermario, improved guide: jajojysazomololu (probably I forgot another 100 people)
|
|
|
|
OkyDooky
|
2023-06-09 01:10:29
|
the French can call it, « J'excelle »
|
|
2023-06-09 01:11:18
|
it would communicate professionalism!
|
|
2023-06-09 05:30:22
|
You use a soft J for your name? \:O
(<@794205442175402004>)
|
|
|
|
_wb_
|
2023-06-09 05:31:20
|
No, it's pronounced like "yon"
|
|
2023-06-09 05:33:21
|
Just like Jyrki is pronounced more like "yeerkee" than like "jerky"
|
|
|
|
yoochan
|
2023-06-09 06:25:49
|
pronounced djerki it would sound like Parkison-son
|
|
|
|
spider-mario
|
|
_wb_
Just like Jyrki is pronounced more like "yeerkee" than like "jerky"
|
|
2023-06-09 06:50:17
|
or maybe even yürkee, no?
|
|
2023-06-09 06:50:57
|
the wiktionary seems to agree https://en.wiktionary.org/wiki/Jyrki#Finnish
|
|
2023-06-09 06:51:05
|
(/y/ in IPA is the ü sound in German)
|
|
2023-06-09 06:51:48
|
https://en.wikipedia.org/wiki/Close_front_rounded_vowel
|
|
2023-06-09 06:52:14
|
ah, or u in Dutch, apparently
|
|
2023-06-09 06:52:55
|
the “Occurrence” section of Wikipedia articles about sounds is often very helpful
|
|
2023-06-09 06:53:07
|
if you can spot a language you know about in there, you can go “aaah, it’s that sound”
|
|
2023-06-09 06:56:58
|
(and conversely, to see the sounds in a given language, you can often try “[language] phonology” and find a good article)
|
|
2023-06-09 06:57:26
|
https://en.wikipedia.org/wiki/Finnish_phonology
|
|
2023-06-09 07:03:21
|
https://en.wikipedia.org/wiki/Finnish_orthography also useful
|
|
2023-06-09 07:04:26
|
y:
> The precise pronunciation tends to be between [y] and [ø].
|
|
2023-06-09 07:04:28
|
uh.
|
|
2023-06-09 07:05:48
|
([ø] is https://en.wikipedia.org/wiki/Close-mid_front_rounded_vowel)
|
|
|
|
_wb_
|
2023-06-09 07:07:16
|
That's the "eu" vowel in Dutch
|
|
2023-06-09 07:07:27
|
So something between Dutch "u" and "eu"
|
|
|
|
Traneptora
|
2023-06-09 07:09:22
|
I always figured Jyrki was like "yur-kee" but I never knew it about Jon
|
|
2023-06-09 07:10:15
|
I always figured it was 'dch' like in the English John
|
|
|
|
_wb_
|
2023-06-09 07:10:54
|
Nah it's like how Jan is pronounced in German or Dutch
|
|
|
|
Traneptora
|
2023-06-09 07:11:24
|
is the last name pronounced "ssneuilleurs"
|
|
2023-06-09 07:11:25
|
(in french)
|
|
2023-06-09 07:12:00
|
I guess I never noticed cause Jon is an english name as-is
|
|
|
|
_wb_
|
2023-06-09 07:17:49
|
Sneyers is pronounced /snɛijərs/
|
|
|
|
OkyDooky
|
2023-06-09 07:19:42
|
That's what I meant by "soft." Like, as apposed to Jon (John, minus the h).
Edit\: It makes no sense that a hard sound is called the "soft" form, especially when I can't find anything on what the "hard" form of a J is. 😠
(<@794205442175402004>)
|
|
|
|
_wb_
|
2023-06-09 07:21:44
|
Sneyers is basically the same name as "Taylors" in English, that is, it means "the son of a tailor". Literally, a "sneyer" (in modern spelling that would be "snijder") is a "cutter", i.e. someone who cuts cloth to make clothes.
|
|
|
|
elfeïn
|
2023-06-09 07:26:03
|
oh wow, that's interesting because I think *tailor* means something similar.
|
|
2023-06-09 07:26:35
|
that sounded a lot more nuanced in my head
|
|
|
|
_wb_
|
|
That's what I meant by "soft." Like, as apposed to Jon (John, minus the h).
Edit\: It makes no sense that a hard sound is called the "soft" form, especially when I can't find anything on what the "hard" form of a J is. 😠
(<@794205442175402004>)
|
|
2023-06-09 07:28:37
|
I know of 3 ways to pronounce a J:
- the "dj" sound (/dʒ/) like in the English John
- the "j" sound (/ʒ/) like in the French "je joue des jeux"
- the "y" consonant (/j/) like how the letter J is used in Dutch or German ("ja")
|
|
2023-06-09 07:33:58
|
Tailor is one of those English words that is just French, tailleur, which is the person who measures your dimensions (taille) in order to cut clothes to size (tailler).
|
|
|
|
Traneptora
|
2023-06-09 07:34:41
|
il est en fraçais originalement
|
|
2023-06-09 07:35:13
|
mais quant Guillaume est arrivé de l'angleterre...
|
|
|
|
_wb_
|
2023-06-09 07:35:30
|
And French got it from Latin
|
|
|
|
Traneptora
|
2023-06-09 07:35:44
|
and Latin got most of its words from old etruscan, iirc
|
|
2023-06-09 07:35:50
|
and its alphabet from ancient greek
|
|
|
|
_wb_
|
2023-06-09 07:35:56
|
Same root as the Italian "tagliatelle"
|
|
2023-06-09 07:36:31
|
Which are of course slices cut from a pasta sheet
|
|
|
|
veluca
|
2023-06-09 07:36:47
|
("tagliare" = "to cut")
|
|
|
|
_wb_
|
2023-06-09 07:37:06
|
So you can call me Jon Tagliatelle if you want
|
|
2023-06-09 07:38:58
|
A "detail" ("un détail") shares the same root even
|
|
|
|
Traneptora
|
2023-06-09 07:39:40
|
~~minor details~~
|
|
2023-06-09 07:39:56
|
I wonder if "tail" does too?
|
|
2023-06-09 07:40:02
|
probably not, since french for that is `queue`
|
|
|
|
OkyDooky
|
2023-06-09 07:42:41
|
Yeah, I didn't think about the French pronunciation being a distinct sound until looking up J, just now. But, that makes sense. Thanks for clarifying. \:)
I'm just salty that I don't actually share my name with the creator of the successor to my favorite image format. Same spelling, though. Lol 😅
(<@794205442175402004>)
|
|
|
|
_wb_
|
2023-06-09 07:43:47
|
Ah lol, you're also called Jon then?
|
|
|
|
OkyDooky
|
2023-06-09 07:43:53
|
Refers to measurement...that makes more sense, now.
(<@794205442175402004>)
|
|
|
|
_wb_
|
2023-06-09 07:45:28
|
Loads of people call me Jon-pronounced-the-John-way, for example almost everyone who doesn't know me
|
|
|
|
OkyDooky
|
2023-06-09 07:45:31
|
Yep! Occassionally, people would mistake it for being short for Jonathan, because of the lack of the "h", but that's just how my parents chose to spell it.
(<@794205442175402004>)
|
|
|
|
_wb_
|
2023-06-09 07:47:24
|
Yeah I also get that sometimes, even occasionally get a letter addressed to Jonathan because someone assumed that would be my official name
|
|
2023-06-09 07:48:01
|
Another funny thing is that my middle initials are actually Ph.D.
|
|
2023-06-09 07:49:07
|
My second name is Philippe (which is also the name of my godfather) and my third name is Dominique (the name of my godmother)
|
|
2023-06-09 07:50:26
|
So I could put Ph.D. in my name even before I got the actual degree 😂
|
|
|
|
OkyDooky
|
2023-06-09 07:50:44
|
Haha, no way. Dealing in any sort of international business seems to guarentee all kinds of entertaining drama like that.
But, still, that's kind of cool that you can claim that. ("Ha. 'Ph.D' is my middle name! No, seriously, it literally is.")
|
|
|
|
Traneptora
|
2023-06-09 07:51:29
|
Jon "Ph. D." Ph. D. Sneyers, Ph. D.
|
|
|
|
OkyDooky
|
2023-06-09 07:51:45
|
It's a great icebreaker, if nothing else. 😄
(<@794205442175402004>)
|
|
|
|
lonjil
|
|
Traneptora
and Latin got most of its words from old etruscan, iirc
|
|
2023-06-09 10:04:08
|
Latin got most of its words from Proto-Indo-European
|
|
|
|
veluca
|
|
Traneptora
and its alphabet from ancient greek
|
|
2023-06-09 10:08:00
|
technically Etruscan alphabet, apparently
|
|
2023-06-09 10:08:59
|
huh I never knew this, and I come from quite close from the Etrurian area
|
|
|
gb82
|
|
_wb_
No, it's pronounced like "yon"
|
|
2023-06-10 02:47:19
|
Oh dang, I'll keep this in mind!
|
|
|
|
fab
|
|
_wb_
Loads of people call me Jon-pronounced-the-John-way, for example almost everyone who doesn't know me
|
|
2023-06-10 08:10:01
|
I pronounce Jon as in English
|
|
|
|
elfeïn
|
2023-06-10 08:11:24
|
huh
|
|
2023-06-10 08:11:39
|
it doesn't even have an E in it
|
|
2023-06-10 08:11:55
|
unless you say Onglish
|
|
|
|
fab
|
2023-06-10 08:18:56
|
Im not accurate at say English
|
|
2023-06-10 08:19:02
|
Max 44%
|
|
2023-06-10 08:19:08
|
In the first sound
|
|
|
|
Traneptora
|
|
veluca
technically Etruscan alphabet, apparently
|
|
2023-06-10 01:34:38
|
which I think is derived from greek
|
|
2023-06-10 01:34:53
|
both come from Phoenician
|
|
|
|
veluca
|
2023-06-10 01:58:14
|
Yup I think so
|
|
|
|
OkyDooky
|
2023-06-10 02:52:31
|
Giovanni Tagliatelle is a great name to give a little persuasion to the uncooperative browsers when you knock on their conferencing room's door with a wrench and a tailored suit 😜
(<@794205442175402004>)
|
|
2023-06-10 02:54:02
|
If that's what it takes to stop them from noodling around…
|
|
|
|
_wb_
|
|
|
veluca
|
2023-06-10 05:05:30
|
doesn't sound particularly southern-italy-ish
|
|
2023-06-10 05:05:33
|
😛
|
|
|
|
DZgas Ж
|
2023-06-11 12:21:06
|
how many percent of the image size changes in lossless compression if rotate the image by 90/180/270 degrees <:JXL:805850130203934781>
|
|
|
yurume
|
2023-06-11 01:32:14
|
without a prior knowledge of the non-rotated image?
|
|
2023-06-11 01:32:41
|
the upper bound is of course 0, by tweaking orientation
|
|
|
|
DZgas Ж
|
|
yurume
the upper bound is of course 0, by tweaking orientation
|
|
2023-06-11 09:04:48
|
0.01%
|
|
|
|
yurume
|
2023-06-11 09:12:48
|
sorry, the upper bound assuming the prior knowledge is 0 (+/- a few bits), I meant
|
|
2023-06-11 09:13:15
|
without that prior knowledge encoders can make suboptimal choices
|
|
|
nathanielcwm
|
2023-06-11 04:26:55
|
Since Debian 12 now has libjxl in their repos I've poked the phabricator feature request for jxl in mediawiki
|
|
2023-06-12 06:43:55
|
And Wikipedia is still on debian 10 💀
|
|
|
Moritz Firsching
|
|
DZgas Ж
how many percent of the image size changes in lossless compression if rotate the image by 90/180/270 degrees <:JXL:805850130203934781>
|
|
2023-06-12 06:52:33
|
There are 8 orientations to consider, since mirroring is also allowed. I tried that a few years ago with photographic content and the savings were less than 1%.
|
|
|
|
DZgas Ж
|
|
|
username
|
2023-06-13 08:51:27
|
I'm working on a update to the "Why JXL" page on jpegxl.info
|
|
2023-06-13 08:51:35
|
before:
|
|
2023-06-13 08:51:41
|
after:
|
|
2023-06-13 08:52:30
|
I couldn't find a better logo for adobe's camera raw program
|
|
2023-06-13 08:56:34
|
oh also maybe it would be a good idea to have the matrix bridge mentioned on the main site? I don't really think spammers/bots would be much of a problem at all like they where before since the spammers/bots where all just reading from a master list of public matrix servers and targeting them all
|
|
2023-06-13 08:57:26
|
bots are unlikely to go through the effort of scanning the web for matrix links
|
|
2023-06-13 08:57:49
|
even then the matrix link could be made into a image instead of raw text if that is a worry
|
|
|
prick
|
2023-06-13 10:46:36
|
There are, I've seen crypto spam bots appear in matrix servers
|
|
2023-06-13 10:47:57
|
The only nice way I can think of to provide a link is writing to the viewer's clipboard when they click a button within some basic rot13 string to avoid a basic scan
|
|
2023-06-13 10:48:13
|
Requires JS but oh well, better than an image
|
|
|
diskorduser
|
|
username
after:
|
|
2023-06-13 01:00:42
|
Can you increase palemoon;s size to match safari and thorium size?
|
|
|
|
username
|
|
diskorduser
Can you increase palemoon;s size to match safari and thorium size?
|
|
2023-06-13 01:06:48
|
I am unsure of how to properly do that because I'm not really a web dev and I kinda only half/somewhat know what I'm doing
|
|
|
|
_wb_
|
2023-06-13 01:20:29
|
you'd need an svg version or higher res raster version of that logo — it currently doesn't upscale the raster images, so if the layout requires it to be bigger than what is available, it will get stuck at that size
|
|
|
|
jonnyawsom3
|
2023-06-13 01:24:34
|
If I were them I'd have given it more contrast too, kinda just looks like a grey circle for me
|
|
|
|
_wb_
|
2023-06-13 01:24:59
|
you mean it looks a bit... pale?
|
|
|
Dzuk
|
|
username
after:
|
|
2023-06-13 03:43:33
|
i dont know if it's relevant to what you're doing or not but it's not just Affinity Photo 2 that accepts JXL but the entire Affinity v2 suite (Designer, Photo, Publisher)
|
|
|
|
username
|
|
Dzuk
i dont know if it's relevant to what you're doing or not but it's not just Affinity Photo 2 that accepts JXL but the entire Affinity v2 suite (Designer, Photo, Publisher)
|
|
2023-06-13 03:47:27
|
I had the feeling it was the whole suite but I wasn't sure how to portray that on the page so I just went the lazy route of just switching out the icon with a proper svg
|
|
2023-06-13 03:47:32
|
https://affinity.serif.com/en-us/press/press-kits/
|
|
2023-06-13 03:47:53
|
I couldn't find anything to represent the whole suite
|
|
|
|
_wb_
|
2023-06-13 03:48:13
|
It's kind of the same with the Adobe suite
|
|
|
|
jonnyawsom3
|
2023-06-13 03:48:51
|
What about this?
https://cdn.serif.com/affinity/img/global/logos/affinity-logo-190920160826.svg
|
|
|
|
_wb_
|
2023-06-13 03:48:53
|
Maybe add the affinity lightroom-like thing too?
|
|
2023-06-13 03:49:44
|
I mean, the two most relevant applications are the photoshop thing and lightroom thing
|
|
2023-06-13 03:50:08
|
The illustrator thing and the other stuff are less relevant for jxl imo
|
|
|
|
jonnyawsom3
|
|
What about this?
https://cdn.serif.com/affinity/img/global/logos/affinity-logo-190920160826.svg
|
|
2023-06-13 03:51:00
|
It's white by default but could work, depends how many icons you want
|
|
|
|
username
|
2023-06-13 03:52:07
|
with the text cut out that could work but I personally think it's a bit too flat looking with it only being a single color
|
|
|
|
_wb_
|
2023-06-13 04:40:02
|
looks like affinity has only 3 programs in their suite, they an the Illustrator-like thing called Designer, a Photoshop-like thing called Photo, and an InDesign-like thing called Publisher
|
|
2023-06-13 04:40:33
|
I would say that Photo is the one that is most relevant for JPEG XL (or for raster images in general)
|
|
2023-06-13 04:41:03
|
the two others may also support jxl import/export, but they're mostly about vector graphics
|
|
|
|
jonnyawsom3
|
2023-06-13 05:00:12
|
Maybe once the splines are good enough ;P
|
|
|
|
spider-mario
|
2023-06-13 06:41:39
|
the other two support JXL too, and it probably makes sense in Publisher as well
|
|
2023-06-13 06:41:47
|
it’s not so uncommon to include raster graphics on a page
|
|
|
|
Traneptora
|
2023-06-13 08:16:50
|
for LF coefficients, what's a "sane default" for the modular stream containing the coeffs?
|
|
|
|
_wb_
|
2023-06-13 08:17:06
|
Yeah I don't have a strong opinion, but if you add Publisher and Designer then probably you should also add Illustrator and InDesign
|
|
|
|
Traneptora
|
2023-06-13 08:17:25
|
currently fixing Gradient in hydrium, not sure if I can do better
|
|
|
|
_wb_
|
|
Traneptora
for LF coefficients, what's a "sane default" for the modular stream containing the coeffs?
|
|
2023-06-13 08:18:01
|
Gradient is pretty good, there's a reason we use it for e1 and e2 lossless
|
|
|
|
Traneptora
|
2023-06-13 08:18:17
|
I'm also considering lossily compressing those, dunno if it makes sense to squeeze them
|
|
|
|
_wb_
|
2023-06-13 08:20:30
|
Squeeze is one way to do lossy, there are some simpler tricks that you could also try like keeping only a few msb of the residuals
|
|
2023-06-13 08:21:00
|
(so you get fewer symbols and entropy coding will work better)
|
|
|
|
username
|
|
_wb_
Yeah I don't have a strong opinion, but if you add Publisher and Designer then probably you should also add Illustrator and InDesign
|
|
2023-06-13 10:42:05
|
opinion on this? I could probably do something similar with Adobe's logos/stuff but I'm unsure how much effort that would be to do properly so I feel like just leaving it like this.
|
|
2023-06-13 10:52:19
|
I feel like keeping it sparse is a good idea since including multiple products from the same company kinda bloats the list, it does feel a bit weird having Apple and Adobe listed 3 times each so I don't want to contribute to that.
|
|
|
|
VcSaJen
|
2023-06-13 11:38:08
|
Cramping logos together does not look good. Also make sure to read logo guidelines from owners.
|
|
|
|
username
|
2023-06-13 11:39:51
|
I couldn't find any guidelines for them: https://affinity.serif.com/en-us/press/press-kits/
|
|
2023-06-13 11:46:30
|
I might just submit the pull request like how it is in this image for the time being: https://discord.com/channels/794206087879852103/794206170445119489/1118100370502197339
|
|
2023-06-13 11:48:02
|
because I'm pretty bad when it comes to making subjective changes or decisions, I end up getting caught on them indefinitely
|
|
|
|
DjangoXL
|
2023-06-14 04:32:30
|
Hello there. Can anyone view jpegxl in nomacs according to this instruction https://jpegxl.io/tutorials/nomacs/ ?
|
|
|
username
opinion on this? I could probably do something similar with Adobe's logos/stuff but I'm unsure how much effort that would be to do properly so I feel like just leaving it like this.
|
|
2023-06-14 04:34:36
|
great work dude. looks simple and clear
|
|
|
|
diskorduser
|
|
DjangoXL
Hello there. Can anyone view jpegxl in nomacs according to this instruction https://jpegxl.io/tutorials/nomacs/ ?
|
|
2023-06-14 06:21:54
|
On windows it doesn't work. Nomacs on linux works
|
|
|
novomesk
|
|
DjangoXL
Hello there. Can anyone view jpegxl in nomacs according to this instruction https://jpegxl.io/tutorials/nomacs/ ?
|
|
2023-06-14 07:02:25
|
That information is obsolete.
Try following installer:
https://github.com/nomacs/nomacs/releases/tag/3.17.2282
|
|
|
|
DjangoXL
|
|
novomesk
That information is obsolete.
Try following installer:
https://github.com/nomacs/nomacs/releases/tag/3.17.2282
|
|
2023-06-14 08:01:08
|
Thx. Beta works fine for me.
|
|
|
|
diskorduser
|
|
novomesk
That information is obsolete.
Try following installer:
https://github.com/nomacs/nomacs/releases/tag/3.17.2282
|
|
2023-06-14 10:20:01
|
So it doesn't work on the latest stable? Right
|
|
|
|
DjangoXL
|
|
|
novomesk
|
|
diskorduser
So it doesn't work on the latest stable? Right
|
|
2023-06-14 12:47:50
|
It could work also with the old version (called stable). Old nomacs was built using Qt 5.14 and when someone put there new plug-in which was built against Qt 5.15, it doesn't work.
It works with old plugin or when someone rebuilds the plugin against Qt 5.14 (so that versions between plugin and nomacs match).
|
|
|
|
_yummersdeluxe_
|
2023-06-14 07:28:47
|
How could I help improve the existing page's design?
|
|
|
derberg
|
|
username
oh also maybe it would be a good idea to have the matrix bridge mentioned on the main site? I don't really think spammers/bots would be much of a problem at all like they where before since the spammers/bots where all just reading from a master list of public matrix servers and targeting them all
|
|
2023-06-14 11:47:22
|
Regarding Matrix:
I have a community that was listed on the joinmatrix list (not anymore since we closed registration on the Matrix server) and in total we got maybe three spammers in a year.
|
|
2023-06-14 11:48:21
|
Discord users are more likely to get spammed (but more via DM) from my experience
|
|
2023-06-14 11:55:10
|
List the Matrix room or better yet: make a space and bridge all other bridge-able channels. The only things that can't be bridged (yet; if a webhook is used) are <#803379415106584626> <#847067365891244042> <#805722506517807104> <#1019989424232222791> <#1019989652079394826>, the VC and threads.
|
|
|
|
VcSaJen
|
2023-06-15 04:20:39
|
Also let guests have read access
|
|
|
|
_wb_
|
2023-06-15 06:20:01
|
We have 1k members on this discord, which apparently is a threshold for being discoverable. Should I enable it?
|
|
|
|
username
|
2023-06-15 06:23:45
|
sounds like it could lead to a bot/spammer problem like what was happening when the matrix bridge was public
|
|
|
|
_wb_
|
2023-06-15 06:38:34
|
I dunno if it would be that much of a problem, you need a discord account for that and bots/spammers on discord can just be banned. On Matrix it's easier for spammers to just create a new account, I think...
|
|
|
|
username
|
2023-06-15 06:39:51
|
I have noticed that bots on discord tend to grab a list of everyone in the server and then start direct messaging them
|
|
|
|
afed
|
2023-06-15 06:41:14
|
i don't think there will be any big problems, discord does a pretty good job with spammers
|
|
|
|
username
|
2023-06-15 06:42:28
|
I kinda forgot that discord added a "message requests" system where you need to accept DMs from random people that try to message you
|
|
2023-06-15 07:00:13
|
oh also in other news I submitted a pull request to the community website: https://github.com/jxl-community/jxl-community.github.io/pull/16
|
|
2023-06-15 07:01:37
|
I decided to put off trying to deal with how the Affinity 2 suite is represented/shown on the site since if I didn't then this pull request would probably be stuck in limbo.
|
|
|
|
_wb_
|
2023-06-15 07:16:55
|
Thanks!
|
|
|
novov
|
|
username
sounds like it could lead to a bot/spammer problem like what was happening when the matrix bridge was public
|
|
2023-06-15 12:01:49
|
I used to be a mod of a server which enabled it, it didn’t cause huge issues
It was quite a niche topic but so is this
|
|
|
|
derberg
|
|
_wb_
We have 1k members on this discord, which apparently is a threshold for being discoverable. Should I enable it?
|
|
2023-06-15 12:17:21
|
Yes
|
|
2023-06-15 12:17:31
|
It doesn't bring much spammers.
|
|
2023-06-15 12:17:44
|
Platforms like disboard bring the spammers (or at least weird people)
|
|
2023-06-15 12:20:15
|
If at any point this would turn into a place that got bots regularly, you could setup a small captcha channel that only new people and a few moderating people can see. That would limit spammy DMs to those.
|
|
2023-06-15 12:20:31
|
There might be better ways to handle that nowadays but well, that's how the biggest places I'm active in still handle it
|
|
|
derberg
Platforms like disboard bring the spammers (or at least weird people)
|
|
2023-06-15 12:21:42
|
To add: usually I get spam DMs from acounts that joined very very big places that I'm only on to be able to use the emotes. My account is on 100 Discord guilds in total (maximum for a free account).
So I would say a small place like this here is pretty much safe from being rolled over by bots.
|
|
|
_wb_
I dunno if it would be that much of a problem, you need a discord account for that and bots/spammers on discord can just be banned. On Matrix it's easier for spammers to just create a new account, I think...
|
|
2023-06-15 12:25:03
|
It should be equally simple tbh if not simpler on Discord*. Unless this guild here requires verfification by phone number.
|
|
2023-06-15 12:26:19
|
*Discord requires no mail address while Matrix has guest accounts in the protocol but most home servers don't have guest sign up (and I think there is at least some limited way to block guests joining a room)
|
|
|
|
lonjil
|
2023-06-15 12:26:21
|
I know some people who used to bridge Matrix to IRC but had to stop because the Matrix side kept getting spam and shit and there were no good ways to deal with it.
|
|
|
w
|
2023-06-15 12:26:47
|
at least on discord there are moderation tools
|
|
|
|
derberg
|
2023-06-15 12:27:23
|
Well, there are moderation tools for Matrix but it would be best to run a home server to make best use of them I guess.
|
|
2023-06-15 12:28:43
|
https://matrix.org/docs/guides/moderation
|
|
2023-06-15 12:30:04
|
Ah, looks like server access control lists can be used without owning the home server the own room is on
|
|
|
username
sounds like it could lead to a bot/spammer problem like what was happening when the matrix bridge was public
|
|
2023-06-15 12:35:50
|
On which home server is the matrix room on btw?
|
|
2023-06-15 12:39:37
|
Ah `#jxl:matrix.org`. Consider migrating the room over to a different home server.
|
|
2023-06-15 12:43:15
|
E.g. opensuse.org (haven't seen spam here), mozilla.org, kde.org, gnome.org or whatever poison you like. Just create an acount there, make that also an admin in the jxl room and then add and then change the main home server address, keeping the old as an alternative way to join for people that know it.
|
|
2023-06-15 12:43:50
|
It's the biggest home server
|
|
2023-06-15 12:49:28
|
Ah, right. That should be checked. I did not really check that yet since I was under the impression those tend to be quite open.
Well, I guess at least mozilla.org should really have no problem here considering what kind of rooms exist. I think opensuse.org can also be talked into tolerating if they don't straight up just allow any kind of sane room.
|
|
2023-06-15 12:51:08
|
Besides that, there is also https://joinmatrix.org/servers
|
|
2023-06-15 12:53:27
|
Well, in the end rooms will get (kinda) mirrored to other home servers the moment users from other home servers join.
|
|
|
|
lonjil
|
2023-06-15 12:56:39
|
Is there any tool that will extract just the DC values of a JXL image? I don't see any djxl option for it.
|
|
2023-06-15 05:17:07
|
I tried decoding a truncated JXL, going smaller until it refused.
|
|
2023-06-15 05:17:53
|
It looks fuzzy and is like, maybe a sixth the total file size?
|
|
2023-06-15 05:31:16
|
7.5x smaller
|
|
2023-06-15 06:31:01
|
I tried the latest git version of djxl and decoding HDR JXL's still results in files that look washed out in everything except Chrome (and presumably very recent mpv). Does anyone know how I can tell whether the ICC profile has the recently merged tone mapping stuff?
|
|
|
|
spider-mario
|
2023-06-15 06:36:03
|
for Rec. 2020 + PQ, the old ICC should be around 9 kB, the new one 4.3 kB
|
|
2023-06-15 06:37:22
|
also, on the old one, `cd-iccdump` will say `Version = 4.3` and show three TRC entries, vs. `Version = 4.4` and an A2B0 tag for the new one
|
|
|
|
lonjil
|
2023-06-15 06:42:28
|
yup, seems to be the new one
|
|
2023-06-15 06:45:18
|
Alright the new one does work in Firefox where the old one didn't
|
|
2023-06-15 06:46:16
|
And Geeqie
|
|
2023-06-15 06:46:55
|
So I guess Qt stuff either doesn't support ICCv4, or doesn't do color management at all >.>
|
|
|
|
spider-mario
|
2023-06-15 06:48:42
|
Qt6 now includes some amount of color management stuff, but you would still have to actually use it
|
|
2023-06-15 06:49:16
|
also, sadly no utility to get the ICC profile associated with the current display (for Windows and X11), nor a way to let macOS know the content colorspace to let it do its thing
|
|
|
|
lonjil
|
2023-06-15 06:49:16
|
Oh, no, Gwenview does work properly even though Dolphin gives an incorrect preview.
|
|
|
|
spider-mario
|
2023-06-15 06:50:14
|
(color management on macOS is handled at the system level: apps tag everything with a colorspace, and macOS does the necessary conversions)
|
|
|
|
lonjil
|
2023-06-15 06:50:28
|
And qimgv which also uses Qt image handling shows it washed out
|
|
|
|
spider-mario
|
2023-06-15 06:51:07
|
neon colors on wide-gamut screens are not a thing on macOS
|
|
2023-06-15 06:51:08
|
it’s quite nice
|
|
|
|
lonjil
|
|
spider-mario
Qt6 now includes some amount of color management stuff, but you would still have to actually use it
|
|
2023-06-15 06:51:17
|
ah, so I suppose Gwenview is the only Qt app I have which is bothering?
|
|
|
|
spider-mario
|
2023-06-15 06:51:26
|
could be
|
|
2023-06-15 06:51:44
|
some KDE apps do try to be diligent about this
|
|
2023-06-15 06:51:52
|
(Krita being probably the most exemplary one)
|
|
|
|
lonjil
|
2023-06-15 06:53:58
|
Tried it in Krita and it looks correct, so that checks out.
|
|
2023-06-15 06:57:55
|
what option would I use to get HLG instead of PQ?
|
|
2023-06-15 07:01:05
|
I tired `RGB_D65_202_Rel_HLG` which seems to have output something at least
|
|
2023-06-15 07:02:29
|
I think it looks a bit less washed out than the PQ version without color management
|
|
|
|
spider-mario
|
2023-06-15 07:13:35
|
if you build libjxl with the devtools enabled, you’ll get a `tools/pq_to_hlg` tool that does it the way that BT.2408 recommends it
|
|
2023-06-15 07:14:52
|
ooh, I don’t think I’ve read the new revision from March 2023, I wonder what’s new
|
|
2023-06-15 07:24:01
|
there seems to be a slight restructuration of section 5 (including SDR content in HDR), and three new annexes
|
|
2023-06-15 07:25:26
|
> Annex 9 - HDR and SDR monitors in close proximity
> Annex 10 - NBCUniversal single-master HDR-SDR workflow
> Annex 11 - Conversion between 203 cd/m² and 100 cd/m² (BT.2035) SDR signal formats
|
|
|
|
lonjil
|
|
spider-mario
if you build libjxl with the devtools enabled, you’ll get a `tools/pq_to_hlg` tool that does it the way that BT.2408 recommends it
|
|
2023-06-15 07:25:33
|
oh yeah that looks even better. Still not as good as you'd want it, but still.
|
|
|
|
fab
|
2023-06-15 08:04:08
|
For medium quality
|
|
2023-06-15 08:04:20
|
Wp2 is 0.160bpp
|
|
2023-06-15 08:04:34
|
AVM is 0.120bpp
|
|
2023-06-15 08:05:08
|
While CJXL is 0.449bpp
|
|
2023-06-15 08:06:17
|
cwp2 -q 50 C:\Users\Use\Pictures\o.png -o C:\Users\Use\Pictures\o.wp2
dwp2 C:\Users\Use\Pictures\o.wp2 -o C:\Users\Use\Pictures\oa.png
cjxl -d 1.154 -e 8 --epf=3 C:\Users\Use\Pictures\o.png C:\Users\Use\Pictures\o.jxl
|
|
2023-06-15 08:08:54
|
cjpegli at d 0.961 is already low quality
|
|
|
|
Traneptora
|
2023-06-15 09:55:49
|
what do people here use for archiving manga scans?
|
|
2023-06-15 09:55:58
|
I'm considering just VarDCT d0.1 e9 gab=0
|
|
2023-06-15 09:56:49
|
I'd prefer to do it lossy as the images are mostly grayscale but not *perfect* grayscale
|
|
2023-06-15 09:57:09
|
like they're RGB but not quite r=g=b for every pixel, just close
|
|
|
|
Quackdoc
|
2023-06-15 09:58:20
|
I personally just do d1 e7, but I mainly do it directly on my phone from tachiyomi DLs
|
|
|
|
Traneptora
|
2023-06-15 09:58:21
|
when I say "archive" I don't mean in the lossless sense, as the source is still obtainable, I just want them on my hard drive for offlineaccess
|
|
|
|
jonnyawsom3
|
2023-06-15 10:26:32
|
Could try modular lossy, or if you stick with VarDCT then d0.3 was usually impercivable to me
|
|
|
Eugene Vert
|
|
Traneptora
what do people here use for archiving manga scans?
|
|
2023-06-15 10:39:13
|
I prefer quantization to 10-12 colors using pngquant and compressing using modular `-j 0 -d 0 -I 1 -E 3 -e 9`
https://discord.com/channels/794206087879852103/794206170445119489/1085316556763955332
|
|
2023-06-15 10:41:48
|
Maybe do some color leveling / apply bilateral filter / jpeg-quantsmooth if scans are not good
|
|
2023-06-15 10:47:10
|
If the scans are not clear at all, `-d 3` or avif
|
|
|
|
Traneptora
|
2023-06-15 10:48:45
|
they're pretty clear and high-res
|
|
2023-06-15 10:48:51
|
no need for stuff like d3
|
|
|
Could try modular lossy, or if you stick with VarDCT then d0.3 was usually impercivable to me
|
|
2023-06-15 10:52:37
|
modular lossy doesn't really work except at extremely low distances
|
|
2023-06-15 10:52:42
|
I found d0.2 to be sorta the threshold
|
|
2023-06-15 10:52:56
|
found d0.5 to be sufficient for my purposes, but I was wondering if there's anything better
|
|
|
Eugene Vert
I prefer quantization to 10-12 colors using pngquant and compressing using modular `-j 0 -d 0 -I 1 -E 3 -e 9`
https://discord.com/channels/794206087879852103/794206170445119489/1085316556763955332
|
|
2023-06-15 10:53:10
|
why lossless after pngquant?
|
|
2023-06-15 10:53:21
|
I guess that could work for the non-color images
|
|
2023-06-15 10:53:41
|
also isn't -m 1 redudant for -d 0
|
|
2023-06-15 10:53:57
|
what does -I 1 -E 3 -j 0 do?
|
|
|
|
spider-mario
|
|
Traneptora
also isn't -m 1 redudant for -d 0
|
|
2023-06-15 10:54:14
|
yes, and in fact I find it risky to mentally associate -m with -d 0
|
|
2023-06-15 10:54:26
|
because then, if you specify just -m, you end up with lossy modular at d1
|
|
2023-06-15 10:54:35
|
ask me how I know
|
|
|
|
Eugene Vert
|
|
Traneptora
I guess that could work for the non-color images
|
|
2023-06-15 10:55:18
|
Yep, I don't know how well this works for colour scans, but for grayscale I've found it to be the best way
|
|
|
|
Quackdoc
|
2023-06-15 10:55:48
|
I never thought of doing that tbh
|
|
2023-06-15 10:55:57
|
I just blanket encode everything with d 1 and call it good lol
|
|
|
|
Eugene Vert
|
|
Traneptora
why lossless after pngquant?
|
|
2023-06-15 11:00:17
|
After reducing the palette to 10-12 colors, lossless usually works better than VarDCT
|
|
|
|
Traneptora
|
2023-06-15 11:05:46
|
what do -I, -E, and -j do?
|
|
|
Fraetor
|
|
Traneptora
what do people here use for archiving manga scans?
|
|
2023-06-15 11:12:23
|
I personally just use `-d 0 --lossless_jpeg=1`, as I find that is good enough for me. I'm not working with pristine sources though.
|
|
|
|
Traneptora
|
2023-06-15 11:14:37
|
these are PNG sources
|
|
|
|
Fraetor
|
2023-06-15 11:15:59
|
I generally work from a mix of sources. The lossless_jpeg option only applies when compressing a jpeg
|
|
2023-06-15 11:17:02
|
You could certainly do better for than me though. I just like that it garentees a saving over the originals.
|
|
2023-06-15 11:19:07
|
This is my script for bulk converting images. You might find it useful with the -r (recursive) option for manga conversion.
|
|
|
|
Traneptora
|
2023-06-15 11:19:37
|
find -exec cjxl?
|
|
2023-06-15 11:19:41
|
why do I need a python script
|
|
|
Demonik
|
2023-06-15 11:22:51
|
can libjxl be built for mac/android (I need a library to decode jxl images)?
|
|
|
|
Traneptora
|
2023-06-15 11:23:18
|
yes and yes
|
|
|
|
Eugene Vert
|
2023-06-15 11:23:51
|
JXL viewer for android: https://github.com/oupson/jxlviewer
|
|
|
|
Fraetor
|
2023-06-15 11:23:53
|
It mostly provides a nicer interface and does some convenient handling for removing the originals only if the encoding suceeds. As well as processing multiple images in parallel.
Was originally a bash script by rewritten in python to better handle edge cases such as weird filenames.
|
|
|
|
Demonik
|
|
Eugene Vert
JXL viewer for android: https://github.com/oupson/jxlviewer
|
|
2023-06-15 11:24:18
|
I don't need viewer, I need library itself for decoding in my code
|
|
2023-06-15 11:24:33
|
but thanks
|
|
2023-06-15 11:25:34
|
ok, I guess I'll have to figure out how to crosscompile it for mac then, thanks
|
|
|
|
Fraetor
|
|
Demonik
can libjxl be built for mac/android (I need a library to decode jxl images)?
|
|
2023-06-15 11:25:37
|
Yes it can. The linked jxlviewer uses it on android.
|
|
|
|
Quackdoc
|
|
Demonik
ok, I guess I'll have to figure out how to crosscompile it for mac then, thanks
|
|
2023-06-15 11:28:53
|
it uses libjxl, and its open source
|
|
2023-06-15 11:28:58
|
use it as a guide/template
|
|
|
|
Demonik
|
2023-06-15 11:29:09
|
yeah, thanks
|
|
|
|
Quackdoc
|
2023-06-15 11:29:12
|
though it's not 1:1 the crosscomp stuff is there
|
|
|
|
Eugene Vert
|
|
Traneptora
what does -I 1 -E 3 -j 0 do?
|
|
2023-06-15 11:30:38
|
`-j` is shortcut for `--lossless_jpeg`
`-I 1 -E 3` -- not sure how it works, but it can sometimes produce results that are faster to encode/decode
|
|
|
|
Traneptora
|
2023-06-15 11:32:05
|
i'll read the docs
|
|
|
|
jonnyawsom3
|
2023-06-15 11:33:20
|
`I` is percentage of pixels to base the MA tree from, `E` is for extra MA tree properties in modular
|
|
|
|
Traneptora
|
2023-06-15 11:34:15
|
looks like -E3 enables properties 16-19
|
|
2023-06-15 11:34:46
|
wouldn't -I1 be 1% of the pixels?
|
|
2023-06-15 11:34:52
|
why would you want that?
|
|
2023-06-15 11:35:05
|
I mean it'd be faster to encode for MA tree learning
|
|
2023-06-15 11:35:11
|
but I don't see how it's better than gradient
|
|
2023-06-15 11:35:25
|
i.e. -P5
|
|
2023-06-15 11:35:49
|
btw, do larger modular groups yield better ratios, or is it somewhat equivalent?
|
|
|
|
jonnyawsom3
|
2023-06-15 11:46:26
|
`I` used to be reversed, so 1 meant full and 0.01 was 1% or something like that. It's worth trying both 1 and 100, sometimes the less complex tree can save space
|
|
|
|
VcSaJen
|
2023-06-16 02:14:51
|
JXL example images need more examples with color profiles: ICCv2 (all features), ICCv4 (all features), color profiles that can be autohandled by library, color profiles that can't be autohandled by library, HDR, wide gamut, etc. Also add textual description of images (png examples can fail if example pngs also use ICCs).
This seems to be the most problematic part in all viewers.
|
|
|
|
_wb_
|
2023-06-16 07:15:41
|
Yeah it would be useful to have an "application test suite" like that
|
|
|
lithium
|
|
Traneptora
what do people here use for archiving manga scans?
|
|
2023-06-16 10:16:27
|
In my opinion, general type grayscale manga content,
-d 1.0 ~ 0.8 -e 7~ 8 will be a stable option,
but for color manga content, I recommend wait next quality improvement,
And some black and white or monochrome manga content will happen DCT worst case,
so still need a lossy palette mode to handle this situation,
(similar av1 palette mode),
or use pngquant + jxl lossless, pingo lossy palette + jxl lossless also a good choose.
|
|
|
|
lonjil
|
|
lonjil
I tried decoding a truncated JXL, going smaller until it refused.
|
|
2023-06-16 10:54:50
|
actually, this is what I figured `--preview_out=FILENAME` would do, but it didn't do anything.
|
|
|
|
Eugene Vert
|
2023-06-16 10:57:48
|
One can also use `cjxl --override_bitdepth=N` for color quantization, but pngquant produces a more consistent palette
|
|
|
|
VcSaJen
|
2023-06-16 12:39:20
|
`cjxl` fails to encode official ICC test images: `browsertest.jpg` and `browsertest_lut.jpg` from https://www.color.org/browsertest.xalter and `Upper_Left.jpg` from https://www.color.org/version4html.xalter. Tried versions `v0.8.1` and `731d3d84`
|
|
2023-06-16 12:40:32
|
Edit: looks like specifically `Upper_Left.jpg` was reported in issue #1810. Not sure if `browsertest.jpg` and `browsertest_lut.jpg` are different case or the same.
|
|
|
|
lonjil
|
2023-06-16 12:46:18
|
fails on the git version I built yesterday so probably not fixed?
|
|
|
|
VcSaJen
|
2023-06-16 01:04:44
|
Commented on that issue.
|
|
|
|
Traneptora
|
2023-06-17 02:17:37
|
looks like if you start with a PNG with alpha, cjpegli just fails with a cryptic error message
|
|
2023-06-17 02:17:51
|
has this already been reported?
|
|
2023-06-17 02:18:32
|
and if not, could it potentially use an Adobe Transform marker to add alpha?
|
|
2023-06-17 02:39:51
|
opened one
|
|
2023-06-17 02:39:52
|
https://github.com/libjxl/libjxl/issues/2579
|
|
2023-06-17 02:50:46
|
Personally, I agree, the cryptic error messages are frustrating
|
|
2023-06-17 02:50:56
|
if there isn't already an issue for that one I'd do it
|
|
2023-06-17 02:52:05
|
it would also help us learn why things like this are failing
|
|
2023-06-17 02:52:06
|
https://github.com/libjxl/libjxl/issues/2542
|
|