SDCONVERT Convert alg into a pipeline or create type conversion pipeline To convert algorithm ALG into a pipeline: OUT=SDCONVERT(ALG) To add type conversion to an existing pipline: P2=SDCONVERT('uint8-double',P) % P2 starts from uint8 data To create only the conversion pipeline including feature selection P=SDCONVERT('uint8-double','input',3,'select',[2 3]) INPUT ALG trained SDALG object OUTPUT OUT converted pipeline OPTIONS 'uint8-double' - convert uint8 into double 'uint16-double' - convert uint16 into double 'input',D - input dimensionality 'select',ind - indices of features to pass through DESCRIPTION SDCONVERT transforms trained object into a pipeline. For example, algorithm ALG may be converted (provided that the algorithm function must has a toconvert section defining the conversion) SDCONVERT can also create a type conversion pipeline which type-casts the input data in uint8 (or uint16) into double. This is useful to create exported pipelines that directly operate on byte/integer image data. By default, specify conversion and provide pipeline object. SDCONVERT will add a conversion step and return the complete pipeline. p=sdfisher(data); pfull=SDCONVERT('uint8-double',p); % pfull can be applied to uint8 buffer Feature selection on the integer inputs may be specified with 'select' option. In this setup, we create a separate conversion pipeline and join later. Input dimensionality must be specified. Example of classifier on G and B bands: >> pc=sdconvert('uint8-double','input',3,'select',2:3) Type conversion pipeline 3x2 uint8-double >> p=sdfisher( data(:,{'G','B'} ) % our data contains R,G,B feature names >> pall=pc*p % complete pipeline starting from R,G,B uint8 image data
sdconvert
is referenced in examples: