Contextual classifications - TrainAutoContext
Purpose
Integration of contextual classification system in iota²
About contextual classifications
prerequisites
SLIC segmentation method has to be operational #158 (closed)
Reference code
The code is available at : https://gitlab.orfeo-toolbox.org/derksend/otb/tree/otb-auto-context/Modules/Remote/ModuleAutoContext
It has been tested on a sample Sentinel-2 image (100x100 pixels of 10 bands).
Observations - limitations
The application has been develop in order to consider a single tile. This issue is about to extend the application's possibilities : from an input of one tile to an image list.
here is the actual application interface :
Parameters:
MISSING -in <string> Input image (mandatory)
MISSING -inseg <string> Input segmentation (mandatory)
MISSING -tmpdir <string> Temporary directory (mandatory)
MISSING -refdata <string> Reference data (mandatory)
MISSING -field <string> Label field (mandatory)
MISSING -out <string> output path (mandatory)
-nit <int32> number of iterations (mandatory, default value is 3)
-ram <int32> Available RAM (MB) (optional, off by default, default value is 256)
-progress <boolean> Report progress
-help <string list> Display long help (empty list), or help for given parameters keys
and what's it could be
Parameters:
MISSING -il <string list> Input image list (mandatory)
MISSING -insegList <string list> Input segmentation list (mandatory)
MISSING -tmpdir <string> Temporary directory (mandatory)
MISSING -refdata <string> Reference data (mandatory)
MISSING -field <string> Label field (mandatory)
MISSING -out <string> output path (mandatory)
-nit <int32> number of iterations (mandatory, default value is 3)
-ram <int32> Available RAM (MB) (optional, off by default, default value is 256)
-progress <boolean> Report progress
-help <string list> Display long help (empty list), or help for given parameters keys
where input image list and input segmented images list must respect the same order. Also, refdata
must be present in all input image.
In order to deal with images a first approach could be to iterate on images until reaching the production of initTrainSamples.shp which is the start of iterations.