Results 1 to 8 of 8

Thread: paq9a modification

  1. #1
    Member
    Join Date
    May 2008
    Location
    Estonia
    Posts
    395
    Thanks
    148
    Thanked 226 Times in 123 Posts

    paq9a modification

    Found this http://ilovemyking.googlepages.com/new77r.cpp
    Has someone tested it?
    KZo


  2. #2
    Moderator

    Join Date
    May 2008
    Location
    Tristan da Cunha
    Posts
    2,034
    Thanks
    0
    Thanked 4 Times in 4 Posts

  3. #3
    Programmer osmanturan's Avatar
    Join Date
    May 2008
    Location
    Mersin, Turkiye
    Posts
    651
    Thanks
    0
    Thanked 0 Times in 0 Posts
    By only looking the code, it's clear it must be very slow. Since it has too much APM stages besides some extra models such DMC and run context map. I think, this work is useless. Since LZP+CM approach is crap for me. You have two choice: encode only a flag for predicted char (you lose some known correlations) or encode a flag for predicted char (you make some extra redundancy) and then update whole statistics with that char (like PAQ9a does). Of course, masking can be implemented. But, proper masking in bit-wise coder could be very expensive. BTW, I wonder this modification efficiency.

  4. #4
    Programmer osmanturan's Avatar
    Join Date
    May 2008
    Location
    Mersin, Turkiye
    Posts
    651
    Thanks
    0
    Thanked 0 Times in 0 Posts
    BTW, here is some information (non-technical though):
    http://ilovemyking.googlepages.com/paqpage

  5. #5
    Programmer osmanturan's Avatar
    Join Date
    May 2008
    Location
    Mersin, Turkiye
    Posts
    651
    Thanks
    0
    Thanked 0 Times in 0 Posts
    SFC Test on my laptop (Core2 Duo 2.2GHz, 2GB RAM, Vista x64 Business SP1)
    Code:
    PAQ9a -7
    a10.jpg      -> 823,940 bytes
    AcroRd32.exe -> 1,235,964 bytes
    english.dic  -> 458,794 bytes
    FlashMX.pdf  -> 3,631,670 bytes
    FP.log       -> 391,475 bytes
    MSO97.dll    -> 1,611,322 bytes
    ohs.doc      -> 727,715 bytes
    rafale.bmp   -> 740,372 bytes
    vcfiu.hlp    -> 492,806 bytes
    world95.txt  -> 431,606 bytes
    Total Size: 10,545,664 bytes
    Total Time: 118.469 Seconds
    
    NEW77R -7
    a10.jpg      -> 824,470 bytes
    AcroRd32.exe -> 1,170,972 bytes
    english.dic  -> 455,927 bytes
    FlashMX.pdf  -> 3,620,117 bytes
    FP.log       -> 391,642 bytes
    MSO97.dll    -> 1,541,976 bytes
    ohs.doc      -> 715,741 bytes
    rafale.bmp   -> 736,290 bytes
    vcfiu.hlp    -> 477,287 bytes
    world95.txt  -> 435,450 bytes
    Total Size: 10,369,872 bytes
    Total Time: 453.981 seconds
    Just ~200kb improvement at cost of ~4x time

  6. #6
    Expert
    Matt Mahoney's Avatar
    Join Date
    May 2008
    Location
    Melbourne, Florida, USA
    Posts
    3,255
    Thanks
    306
    Thanked 779 Times in 486 Posts
    The program doesn't say, but from the website it must be Bill Pettis, author of paq8k (a very slow version of paq8j with lots more models). I think to squeeze more compression out of it, you need to drop the LZP preprocessor. I added it mainly as a speed optimization but LZP doesn't mix well with CM.

  7. #7
    Member chornobyl's Avatar
    Join Date
    May 2008
    Location
    ua/kiev
    Posts
    153
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Speeding up arithmetic coding

    Sorry for offtop but i found this link, may it be useful for someone of you

    http://diracvideo.org/download/arith...h-speedups.pdf

  8. #8
    Administrator Shelwien's Avatar
    Join Date
    May 2008
    Location
    Kharkov, Ukraine
    Posts
    3,271
    Thanks
    201
    Thanked 985 Times in 511 Posts
    That's worth reading, but is rather obvious and related purely to their own coder.
    Also I suspect that despite these optimizations, their coder would work
    slower than, say, Matt's rc, as its more complex.
    And 8 bits per probability is not enough so there'd be redundancy, which
    slows down the coder even more (as its renormalization has to process
    more bytes).

    So I'd recommend checking out fpaq0pv4B instead
    http://ctxmodel.net/files/fpaq0pv4b2.rar

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •