Here are some benchmarks. Stuffit is still the fastest (even in "slow" mode) and gets the best compression. I could not time it exactly because it uses a GUI. The other times are process times averaged over several runs on a 2.2 GHz Athlon-64 3500+, 2 GB memory, WinXP SP2 32 bit. I didn't time decompression.
716,606 a10.pjg PackJPG 2.0 (2.8 sec)
698,510 a10.jpg.paq8l (21.5 sec)
698,214 a10.jpg.paq8f (17.6 sec)
697,818 a10.pjg PackJPG 2.2 (1.58 sec)
674,995 a10.jpg.paq8m (33.8 sec)
667,190 a10.jpg.paq8fthis (44.0 sec)
638,541 a10.jpg.sitx stuffit 9.5 (from maximumcompession.com)
638,540 a10.jpg.sitx stuffit 11 (< 1 sec)
so packjpg has chance to compete with stuffit.. but i have tested v2.2 agains large jpg file ... its too slow .. to speed it up the jpg coeffiecient has to be divided into block and compress each block alone.. even more if packjpg support recent processors (sse3 due core or quad core) it would send each block to a thread then gathered into a single data stream..
take a look at: http://www.compressionmax.com/jpg.html . packjpg easily outperforms stuffit on this test.
i guess that stuffit uses and predefined (fixed) model for compressing jpegs, while paq and packjpg adapts for encoded data.
this can explain stuffit efficiency on usual images and inefficiency on unusual images.
maybe it's possible for someone to gather few hundreds of megabytes of jpegs of different sizes, noise levels, blurring, compression level, etc, make some model that will learn from that data and then freeze the model and create normal jpeg packer on it.
this way we can beat stuffit
good but what is the sample.. its a large on (11mb) .. but how much is the time difference,, stuffit is around 4 times faster.. but 14% worse.. so your result second PackJpg potential.. thus future
what? stuffit is only 70 % faster on this sample.
packjpg should certainly have some prebuilt model (btw: durilca had some), that should help compressing eg. first megabyte of jpeg data. prebuilt model is a secret of stuffit.
sorry my mistake ... but what is this image ?
something unusual i think. maybe like that: http://i10.tinypic.com/6cewi1k.jpg
some images can be compressed better by using lempel ziv algorithm instead of huffman - if they contain lots of <u>identical</u> blocks. stuffit in jpeg mode doesn't look for identical (or similar) blocks across the entire image.
identical blocks appear if the image is artifical. for example many people uses jpeg to compress image charts with lots of text. i know that png or even simple deflate compresses a lot better than jpeg on such files, but some people are just stupid or stubborn
i meant in the test .. its 11.4 mb is it a nature image or .. becuase it compress well .. that i think its rather a bad encoded jpg as good encoded jpg usually compress less .. i have test my self.. and the result indicate that the jpg settings used to save this image don't suite image data type..
Maybe not that bad, but I guess you're right, it is an artificial image.
19,693 6cewi1k.sitx (Stuffit 11, smallest JPEG, no thumbnails)
2,581 6cewi1k.pjg (packJPG 2.2)
1,323 6cewi1k.jpg.zip (zip -9)
960 6cewi1k.pmd (ppmd no options)
816 6cdwi1k.jpg.sr2 (sr2 no options)
661 6cewi1k.pmm (ppmonstr no options)
568 6cewi1k.jpg.paq8fthis (-6)
563 6cdwi1k.jpg.lpaq1 (6)
482 6cewi1k.jpg.paq8l (-6)
482 6cewi1k.jpg.paq8m (-6)
59,683 6cewi1k-opt.jpg (jpegtran -optimize)
did some one managed to make packjpg a preprocessor and to like for jpgs inside files..like (doc,xls,mov,rpc, jng)
office files tend to have images, and one using the compress image feature it convert all to jpg thus its required
quicktime movies captured with digital camera (for example Panasonic) and AVI (other models) are using JPG with lower resolution to capture motion.. thus its required
Archivision Files (*.rpc) http://www.archvision.com/ContentCategories.cfm?Ca tegory=1
are images of opjects or people taken around them 360 degree and used for rendering (max,maya,rhino,autocad,...)
then JNG which is an extension to MNG format (the animated brother of PNG) with JPG algorithem (i can't recall the version used).. to be in mind..
so in short.. i have tested Paq8M but does not succeed in getting all of it.. and it still not practical today (maybe in future.. even then a more powerful borther would appear)
so my idea is to get jpg compressor to compress file2file mode or in the archivision case files2file because each frame has a much similar data thus there is a big chance to match coeffiecents..or lets call it Merging Coeffiecent..
may be for now but the precomp + packjpg has to be filters or preprocessor to any archiver..(of course based on theor license)
what is in my mind is LZX lossless unpacker for (chm,chw,wmi,cab,lit,..) then it would be perfect...
precomp works on zlib streams but still don't know other deflates like MSZIP in cab, Deflate64, Deflate in old installers (wise..,.)
CHM files has inside
html >> text compression
gif >> precomp
png >> precomp
jpg >> packjpg
you can take alook at any CHM file in windows with 7-zip and it show the LZX setting used (note that maximum LZX=21).. maximum one i found is 17... even you can unpack it with 7-zip to folder.. compress that folder with WinRAR or 7-zip and see how much you can save...
i think the future work to be focusd on these preprocessor
If anybody wants latest version of StuffIt Deluxe (ONLY for tests, of course!), ask me in PM.
Oooops, this forum doesn't have PM
Yeah, I would appreciate such thing, since I have no other way to get it for my tests
You can provide your e-mail, i'll send you link
Thank you very much!
i had an old test version of a chm optimizer issued by Igor Pavlov.. but i have to ask him before i post it here..
are you serious?Originally Posted by Surfer
Iirc this is a forum dealing with compression-questions and no filesharig-club for software thats considered pirate-copy in some countries!
If anyone wants to deal with that, find other way to exchange that!
Sorry, but I do not want to support that; what anyone does at home is his personal thing, but leave it there. Please!!!
are you asked them officially? are they even dont provide trial versions??Originally Posted by Black_Fox
Some time ago I got freely the trial version of StuffIt 11 from stuffit.com.Originally Posted by Bulat Ziganshin
No official request was made by me, since some test results are not released, when I get to next benchmark release someday, I of course ask Smith Micro prior to that (when not thinking about legal issues, its polite at least).Originally Posted by Bulat Ziganshin
Trial version are provided in exchange for credit card data - at least when I tried that last time. I dont have credit card.
Matt Mahoney has had an experience with StuffIt trial versions (read the note)
Thanks, encode. Nevertheless, the file there is the exactly same I've got mailed.
I thought that Deluxe version has better compression (like improved parsing) over Standard, but it hasn't... that solves all problems introduced. Sorry.
The rules here do not currently forbid the posting of illegal/pirate related material. But as moderator I would strongly advise against such behaviour for the sake of this forums reputation if nothing else.Originally Posted by Vacon
There are some news on the packjpg website (for everyone who didn?t see it yet). There is a MacOS version and a paper which describes the algorithm / technique behind packjpg (very well written!)
I missed 2.3a completely...Originally Posted by Simon Berger
Thank you for pointing us there!
Yes the "a" version fixed many crashes with corrupted or pictures which don?t have the absolutly right format. I found many of them in zipped game engine packages of medal of honor and enemy territory while playing with precomp.