DataCompressor usage

DataCompressor allows compression of image files using scalar and vector quantization.

The application can load RAW image files and all formats supported by the SCIFIO library.

Help output:

usage: azgracompress.DataCompressor [options] input
 -b,--bits <arg>                   Bit count per pixel [Default 8]
 -bench,--benchmark                Benchmark
 -c,--compress                     Compress 16 bit raw image
 -cbc,--codebook-cache <arg>       Folder of codebook caches
 -d,--decompress                   Decompress 16 bit raw image
 -h,--help                         Print help
 -i,--inspect                      Inspect the compressed file
 -mp,--middle-plane                Use middle plane for codebook creation
 -o,--output <arg>                 Custom output file
 -sq,--scalar-quantization         Use scalar quantization.
 -tcb,--train-codebook             Train codebook and save learned
                                   codebook to cache file.
 -v,--verbose                      Make program verbose
 -vq,--vector-quantization <arg>   Use vector quantization. Need to pass
                                   vector size eg. 9,9x1,3x3
 -wc,--worker-count <arg>          Number of worker threads

Quantization types (QT):

• This program supports two different quantization types:
• Scalar quantization, selected by -sq or --scalar-quantization
• Vector quantization, selected by -vq or --vector-quantization
  • Vector quantization requires you to input the vector dimension after the flag
  • For one-dimensional row vectors you can the length as 9 or 9x1
  • For two-dimensional matrix vectors the dimensions is set by DxD format, eg. 3x3, 5x3
  • For three-dimensional voxel vectors the dimensions is set by DxDxD format, eg. 3x3x3, 5x3x2

Main program methods

Compress

• Use with -c or --compress
• Compress the selected image planes (Currently supporting only loading from RAW image files).
• QT is required
• Set the bits per pixel using -b or --bits and integer value from 1 to 8. Codebook size is equal to (2^bits).
• Normally the codebook is created for each image plane separately, if one wants to use general codebook for all planes, these are the options:
  • Set the middle plane index using -mp or --middle-plane. Middle plane is used to create codebook for all planes.
  • Set the cache folder by cbc or --codebook-cache, quantizer will look for cached codebook of given file and codebook size.
• For input file info see Input File section

Decompress

• Use with -d or --decompress
• Decompress the file compressed by this application.
• This method doesn't require any additional options.

Inspect

• Use with -i or --inspect
• Inspect the compressed image file or cached codebook.
• Read compressed file header are write the information from that header.

Train codebook

• Use with -tcb or --train-codebook
• QT is required
• This method load all the selected input planes and create one codebook.
• Codebook is saved to the cache folder configured by the -o option.
• Codebook is trained from planes configured by the input file, see Input File section

Benchmark

• Use with -bench or --benchmark
• QT is required.
• Run benchmarking code on input planes with selected quantization type,

Input file

• Input file is required for all methods.
• Decompress and inspect require only the input file path, while other also require its dimensions
• If the input file is RAW data file, then image dimensions must be provided as follows:
  • Input file dimensions are inputed in format of DxDxD [D] [D-D]
    • DxDxD is image dimension. Eg. 1920x1080x1, 1041x996x946 (946 planes of 1041x996 images)
    • [D] is optional plane index. Only this plane will be compressed.
    • [D-D] is optional plane range. Only plane in this range will be compressed.
  • D stands for integer values.
• Planes selected by the index or plane range are used for:
  • Compression
  • Training of codebook
  • Running benchmark

Additional options:

• -v, --verbose - Make program output verbose.
• -o, --output - Set the ouput of compression, decompression, codebook training, benchmark.
• -wc, --worker-count - Set the number of worker threads.

GitHub mirror link