Paq8o3

[squxe]

From MC guestbook:

KZ:
PAQ8o3
http://www.450mb.com/quickupload/20_1189492301.zip
has new jpeg model and PGM model, that all ... only source,(works for me)

Can anybody compile it?

[nimdamsk]

http://www.450mb.com/quickupload/36_1189522310.exe

Didn't manage to compile with g++. Compiled with VC Express /O2.

[LovePimple]

Thanks Squxe, and nimdamsk!

Mirror: Download

[kaitz]

Don't know if right place for discuss it but, when i made PAQ8o3
i looked the PGM graysceale model i got idea that it can made to compress 8 bit bmp uncompressed files to. As the data is same, i mean grayscale =bitmap where bitmap has pallete as grayscale has to. I can be wrong about this of corce.

EDIT:
Made quick test and only speed is faster compression is allmoust same. bmp is bottom up stuff

[LovePimple]

Thanks KZo! Welcome to the forum!

[Black_Fox1]

Hi KZo and welcome to the forums!

Do you mean that the new model compressed 8-bit palette bitmaps as well as the grayscale bitmaps were compressed by PGM model? Sounds promising!

[kaitz]

Hi KZo and welcome to the forums!

Do you mean that the new model compressed 8-bit palette bitmaps as well as the grayscale bitmaps were compressed by PGM model? Sounds promising!

i made bmp 256color image then compressed it.(got size)

then i made pgm image form 256color image, coping header from pgm image and then image data form 256color image. all are ok.
its a hack.
so then compressed so called hacked image and speed is as it was pgm image. compressed output is allmoust same.(cmp img sizes and speed)

but it can easy add to pgm model(and i whant do do it ), only dif is parese other type image header.

hope i cleared out my idea.

EDIT:
Image is jpg converted wiht irfanview to 256 color image.
File SizeIn SizeOut Time
test.bmp 342078 108409 51.10 sec paq8o3 new pgm
test.bmp 342078 113827 189.99 sec paq8o3

[Piotr Tarsa]

File SizeIn SizeOut Time
test.bmp 342078 108409 51.10 sec paq8o3 new pgm
test.bmp 342078 113827 189.99 sec paq8o3

first one has much better efficiency. 5 % smaller file and 4 times faster!!!

you mean that first one is color image data contained in grayscale pgm file format? if so, your hack is good

[kaitz]

you mean that first one is color image data contained in grayscale pgm file format? if so, your hack is good

no first test whas hack, to test can it be done.

File SizeIn SizeOut Time
test.bmp 342078 108409 51.10 sec paq8o3 new pgm
test.bmp 342078 113827 189.99 sec paq8o3

it is PGM model with ability to compress 8 bit (256 color) bmp image.
As pgm image data = 256 color bmp imaga data

[Piotr Tarsa]

still not clever for me

hacked version compresses better?

[kaitz]

hacked version compresses better?

hacked version conteined pgm image header and 8 bit image data(not original pgm image data). so i dont modify image model olny image. and it compressed faster.
after that i aded ability to pgm model to detect bmp 8 bit image and use its image data.

my native is not english so..

EDIT:
another test
File SizeIn SizeOut Time
airplane11280.bmp 1311798 310194 265.86 sec PAQ8o3 new pgm model
airplane11280.bmp 1311798 324334 1089.63 sec PAQ8o3

made SVN at Google. So i can better track changes in paq8

http://code.google.com/p/paq8/source

new model is uploaded there

[squxe]

What about win32 EXE compile?

[kaitz]

What about win32 EXE compile?

http://paq8.googlecode.com/svn/trunk/paq8o4.exe
its intel compiler compiled

[squxe]

thank you KZo!

[LovePimple]

Thanks KZo!

Mirror: Download

[Matt Mahoney]

http://paq8.googlecode.com/svn/trunk/paq8o4.exe
its intel compiler compiled

I tested with calgary.tar and enwik8. Compression is identical to versions back to paq8l. The Intel compile is 8% faster than g++

Directory traversal and creation doesnt work. You need to compile with -DWINDOWS

The paq8o4.cpp source was changed to old style "for" loop scoping rules, so it wont work with up to date compilers. Must be an old version of the Intel compiler. VC++ had this problem until recently.

[kaitz]

The paq8o4.cpp source was changed to old style "for" loop scoping rules, so it wont work with up to date compilers.

I will fix that, when i have time.

[darkbasic]

Hi! I cannot compile paq8o4 with g++
I'm downloading Intel Compiler, but how to link paq7asm-x86_64.asm with intel compiler?

[Matt Mahoney]

I have posted paq8o4 v2 to http://cs.fit.edu/~mmahoney/compression/
It is archive compatible with the version by KZ, so it gets a new version number, not a new name. I modified the source so it should compile with either the old or new "for" loop scoping rules (I hope). I included a g++ executable that should work with wildcards, directory traversal and directory creation like the earlier versions (but slower than the Intel version).

[kaitz]

I have posted paq8o4 v2 to http://cs.fit.edu/~mmahoney/compression/
It is archive compatible with the version by KZ, so it gets a new version number, not a new name. I modified the source so it should compile with either the old or new "for" loop scoping rules (I hope). I included a g++ executable that should work with wildcards, directory traversal and directory creation like the earlier versions (but slower than the Intel version).

did not look before SVN update. So now is in SVN paq8o4 v2 from http://cs.fit.edu/~mmahoney/compression/
added paq8o4v2.zip file in download section in http://code.google.com/p/paq8/downloads/list

[LovePimple]

I have posted paq8o4 v2 to http://cs.fit.edu/~mmahoney/compression/
It is archive compatible with the version by KZ, so it gets a new version number, not a new name. I modified the source so it should compile with either the old or new "for" loop scoping rules (I hope). I included a g++ executable that should work with wildcards, directory traversal and directory creation like the earlier versions (but slower than the Intel version).

Thanks Matt!

Mirror: Download

[darkbasic]

Thank you!
How to compress from the standard input (to the stdout) or from a named pipe?

[Matt Mahoney]

It is easy in Linux. In Windows you have to reset stdin and stdout to binary mode. There is an example in jpg2dct.c but it only works in g++
http://cs.fit.edu/~mmahoney/compression/#jpeg

[kaitz]

File_____SizeIn___SizeOut_Time _____MEM
lena.bmp 786486 > 410614 44.03 sec, 260885722 paq8o4
lena.bmp 786486 > 409913 47.96 sec, 260886770 paq8o4+mod

is it worth to look more and harder

[darkbasic]

@Matt Mahoney
I'm not a developer
Con you please help me patching the source code?
What about adding this feature in the main tree?

[kaitz]

made new vesion from paq8o4 and lpaq1. added APM and StateMap from lpaq1.
Nothing else changed.
Hope it is what Matt Mahoney ment.

Results:
option used -6
File______SizeIn_____SizeOut__Time_MEM
calgary.tar 3152896 594738 1752.92 sec 499256071 paq8o4+new APM, StateMap
calgary.tar 3152896 595533 1606.39 sec 487378182 paq8o4
EDIT:
Binary for testing only:
http://paq8.googlecode.com/svn/trunk/paq8o5.exe

EDIT2:
option used -7
File______SizeIn___SizeOut_Time_______MEM
calgary.tar 3152896 594048 440.55 sec, 885132047 paq8o4+new APM, StateMap

speed is faster becose PC is diffrent(AMD Athlon 64 3500+)

[kaitz]

option used -6
File__________SizeIn___SizeOut _Time_____MEM
DSCN3974.jpg 1114198 834677 52.11 sec 509911213 paq8o4+new APM, StateMap
SVN .exe version is not updated for this result

[LovePimple]

@KZo

Did you compile PAQ8O5 with -DWINDOWS as Matt suggested?

Directory traversal and creation doesnt work. You need to compile with -DWINDOWS

The paq8o4.cpp source was changed to old style "for" loop scoping rules, so it wont work with up to date compilers. Must be an old version of the Intel compiler. VC++ had this problem until recently.

[kaitz]

@KZo

Did you compile PAQ8O5 with -DWINDOWS as Matt suggested?

deem
PAQ8O5 is only initial testing not for public usage.
but i remember DWINDOWS

[LovePimple]

Why did you not compile PAQ8O5 with "-DWINDOWS"?

Will you be providing us with another PAQ8O5 executable compiled with "-DWINDOWS" as suggested by Matt Mahoney?