GMIC binary floating point output file format

View unanswered posts | View active topics

Board index » Gimp Scripts and Plugins » G'MIC Discussions

All times are UTC - 5 hours [ DST ]

Page 1 of 1

[ 5 posts ]

Print view

Previous topic | Next topic

Author Message

apixosoft

Post subject: GMIC binary floating point output file format

Posted: Mon Mar 21, 2016 4:05 am (#1)

New Member

Joined: Mar 21, 2016
Posts: 3

Hi!
I'm trying to work with binary output from GMIC, but cannot find any information about its format. My researches led me to following conclusions ↓
File consists of main header and data blocks which consist of block header and block data. Main header is

Code: Select all

N <pixel type> <byte order>\0x0A

where N is a number of block entries (as I suppose), <pixel type> is float and <byte order> is little_endian.
Next to it is a block header

Code: Select all

W H BpC NoC #S\0x0A

where W is a picture width, H is a picture height, BpC is byte-per-channel, NoC is a number of channels, and S is the size of data which follows this header. But what I see then cannot be described!

As written in GMIC documentation, calculations and data storage is made using 32-bit floats. Ok, let's assume the following command line

Code: Select all

GMIC RGB.png -luminance --fft -append[-2,-1] c -norm. -log. -shift. 50%,50%,0,0,2 -o Float

for 5×5 RGB image. On output we have this ↓

There are 3 image blocks: two 5x5 1-channel 8-bit images and strange 1×21 1-channel 8-bit image. But data sizes of 38 and 70 cannot be divided to 5×5=25! For 32-bit float data size must be 5×5×4=100 bytes.

What is a magical format used to store pixel data?

Top

Ronounours

Post subject: Re: GMIC binary floating point output file format

Posted: Mon Mar 21, 2016 6:32 am (#2)

GimpChat Member

Joined: Jan 03, 2011
Posts: 1656

Hi,

G'MIC uses by default the .cimgz file format, which is basically a file format defined specifically to store data in the CImg Library
(C++ template image processing library : http://cimg.eu).
The format is not that simple to read, as the image data are compressed (but in a non-destructive way). It uses the zlib functions to
compress/uncompress pixel buffers.
If you want more details about how it can be read/saved, you should look at the CImg source code, specifically the methods
CImgList<T>::_load_cimg() and CImgList<T>::_save_cimg() in file CImg.h where all the details of the implementation can be found.
CImg.h can be found online here : https://raw.githubusercontent.com/dtsch ... ter/CImg.h

If you are allowed to write C++ code, then of course, I'd recommend to directly use CImg, it becomes very simple to use, for instance :

Code: Select all

#define cimg_use_zlib
#include "CImg.h"
using namespace cimg_library;
int main() {
  CImg<> img("file.cimgz);
  img.blur(10);
  img.save("foo.cimgz"); 
  return 0;
}

Top

apixosoft

Post subject: Re: GMIC binary floating point output file format

Posted: Mon Mar 21, 2016 6:45 am (#3)

New Member

Joined: Mar 21, 2016
Posts: 3

So the data is compressed, I should supposet it…
Ok, thanks. If I won't find another way to export NORMAL pictures I use that library. BTW, how to make GMIC to export pictures like in preview screen? When specifying .png as an export extension, it exports almost black picture (because it saves colors rounded to integer, not scaling its values to 255). Just compare pictures below (you'll need to move the white point on "black" image according to histogram).

Attachments: