packJPG v2.5C3

[packDEV]

Hi!


I'm Matthias Stirner, the developer of packJPG. I'm going to open-source packJPG within the next few weeks, and I figured it couldn't hurt to post the release candidate here for those who want to try it. The final version will be released along the source code.

I already fixed a few bugs, but maybe there are still some other files that can lead to a crash (with the old v2.4, too). If so, please post here or contact me.


Before I open-source, I still need to do some code cleanup, renew the docs, and, most importantly, to settle for a license. I'd go for the GPL, but then I also want to give freeware closed source developers a no-hassles chance to use the library. If someone makes money from my work, however, I want a share of it or at least be asked for permission.

My ideas so far involve dual licensing (GPL+???) or GPL + special permissions for closed source developers. Any ideas via PN or email would be very welcome.

[Piotr Tarsa]

QuickLZ is dual licensed, you can contact the author.

[Intrinsic]

Minor issue but it didn't like me using /v2 as a switch, had to use -v2.

And hurrah -dev is working, i see -c is now -t. Although the program crashed on every file(large or small) i tried when using -s50 to -s51 & -t1 to -t10, ie packjpg -dev -s50 -t1 {file}
It just stops at:
Processing file 1 of 1 [ ]
with exe has encountered a problem and needs to close, on small files the size sits at 0 bytes but on larger ones it was 8192 bytes. This is on XP SP3.
Oddly -t0 worked though.

So far it looks like some nice gains on files, it's still not so good with optimised or progressive optimised jpegs compared to stuffit though.

[packDEV]

Thanks for pointing this out! Well, I didn't pay attention . The actual maximum for the segmentation setting (s segments) is 49. The setting is clamped internally now, so if you choose, say -s50, it would just use -s49 instead. Same goes for -s0 which would use -s1 instead.

-t? controls the noise threshold, that is, for each coefficient, the upper t bitplanes are considered non-noise, everything below is considered noise and coded differently with a simpler approach. If you choose -t0, everything is considered noise, and everything is coded the same. Similarily, if you choose -t10, nothing is considered noise, and everything is coded the same, too (but differently). The optimum for a given file lies in between. Note that, due to quantization, coefficients in the 64 spectral bands are typically made up of something between 6 and 8 bitplanes, so you might not see any difference between -t10 and -t9.

Btw, the default for segmentation (-s?) is 10. The default for the noise threshold (-t?) is dependant on the size of the input file. Higher sizes mean higher t - theres more data for the statistical model to learn in bigger files, so less has to be considered noise.

The inclusion of the developer options is controlled by a simple define in the code. If someone had asked me I would have been happy to make a developer-enabled v2.4, too .