HELP.txt

SYNOPSIS:
  ./vlg_extractor [COMMANDS] [OPTIONS]  list_input  input_dir  output_dir

  list_input: This is an ascii-file containing the relative path
              (to input_dir) of the input files
  input_dir:  Directory that contains the images
  output_dir: Directory that will contain the features

COMMANDS:
  --help: It shows this help

  --extract_classemes=[OUTPUT_TYPE]
        This extract the continuous version
  --extract_classemes_bin=[OUTPUT_TYPE]
        This extract the binary version
  --extract_picodes128=[OUTPUT_TYPE]
  --extract_picodes1024=[OUTPUT_TYPE]
  --extract_picodes2048=[OUTPUT_TYPE]
  --extract_mc=[OUTPUT_TYPE]
  --extract_mc_bit=[OUTPUT_TYPE]
  --extract_concatenationLowLevelFeatures=[OUTPUT_TYPE]


OPTIONS:
  --parameters-dir=<path directory>
       The directory of the parameters. Default is "<root>/data".

  --classeme-suffix
       This specifies the suffix to use for the classemes output files.
       Default "_classemes".
       Note: in case of "--extract_classemes_bin", we append a "BIN" at the end
             of the filename.

  --picodes-suffix
       Default: "_picodes"
       Note A proper number indicating the dimensionality
       will be append to the filename

  --mc-suffix
      Default: "_mc"

  --load-concatenation-low-level-features=[INPUT_TYPE]

  --concatenation-low-level-features-prefix
      Default: "_low_level_features_v2"

  --disable-output-overwriting
      It does not overwrite a file with the same name on disk
      (by default, the software overwrites everything).


[OUTPUT_TYPE]
   It can be {ASCII, FLOAT, BIN}
     ASCII is an ascii file containing a matrix.
           The software automatically adds extension ".ascii" to this filename.
     FLOAT is a binary file, containing a matrix. The values are stored in
           row order.
           The format is: <uint: n_rows><uint: n_cols><float: data[0]>...
                             <float: data[n_rows*n_cols-1]>
		       where uint and float are 4-bytes each.
           The software automatically adds extension ".dat" to this filename.
           We provide function "load_float_matrix.m" to load this type of files
           into Matlab.
      BIN  This produces the most compact output file: it saves a file which
	       i-th bit corresponds to i-th bit entry.
		   The unused entries in the last byte are filled with zeros.
           The extension is ".bin".
           We provide function "load_bit_vector.m" to load this type of files
           into Matlab.
           Note that this format makes sense only for "classemes_bin",
           "picodes" and "mc_bit".

      Note: there can be multiple [OUTPUT_TYPE] options in order to produce
            more output.

[INPUT_TYPE]
   It can be FLOAT. See the [OUTPUT_TYPE] description above.


NOTES about extract_concatenationLowLevelFeatures:
  This command will extract the low-level features used for the
  construction of the high-level representation classemes, picodes, mc and
  their binary variants. The resulting vector consisting of the
  concatenation of such features has 22860 dimensions, organized as follows:
  - GIST level 0: dims 1:960
  - PHOG-180 level 0: dims 961,980
             level 1: dims 981,1060
             level 2: dims 1061,1380
             level 3: dims 1381,2660
  - PHOG-360 level 0: dims 2661,2700
             level 1: dims 2701,2860
             level 2: dims 2861,3500
             level 3: dims 3501,6060
  - SSIM level 0: dims 6061,6360
         level 1: dims 6361,7560
         level 2: dims 7561,12360
  - SIFT-BOW level 0: dims 12361,12860
             level 1: dims 12861,14860
             level 2: dims 14861,22860

AUTHORS:
  * Alessandro Bergamo
    aleb@cs.dartmouth.edu

  * Chen Fang
    chen@cs.dartmouth.edu

  * Lorenzo Torresani
    lorenzo@cs.dartmouth.edu

REPORTING BUGS:
  Report bugs to Alessandro Bergamo aleb@cs.dartmouth.edu
  Please send a precise and complete report.

REFERENCES:
  [1] Efficient Object Category Recognition using Classemes,
      Lorenzo Torresani et others, ECCV2010
  [2] Alessandro Bergamo, Lorenzo Torresani, Andrew Fitzgibbon
      PiCoDes: Learning a Compact Code for Novel-Category Recognition, NIPS 2011
  [3] Alessandro Bergamo, Lorenzo Torresani
      Meta-Class Features for Large-Scale Object Categorization on a Budget