How to write an application

This chapter presents the different steps to write your own application, and the framework surrounding it.

Application design

The first logical step is to define the role of your application:

  • What is the function of your application? Try to draw a box diagram to describe the design of your application. Note that you don’t have to worry about opening and saving images (or vector data) files, this is handled by the framework.

  • What variables (or data objects) must be exposed outside the application? Try to make a list of the inputs, outputs and parameters of your application.

Then you should have a good vision of your application pipeline. Depending on the different filters used, the application can be streamed and threaded. The threading capabilities can be different between the filters so there is no overall threading parameter (by default, each filter has its own threading settings).

It is a different story for streaming. Since the image writers are handled within the framework and outside the reach of the developer, the default behaviour is to use streaming. If one of the filters doesn’t support streaming, it will enlarge the requested output region to the largest possible region and the entire image will be processed at once. As a result, the developer doesn’t have to handle streaming nor threading. However, there is a way to choose the number of streaming divisions (see section Parameters selection).

Architecture of the class

Every application derives from the class otb::Wrapper::Application. An application can’t be templated. It must contain the standard class typedefs and a call to the OTB_APPLICATION_EXPORT macro.

You need also to define standard macros itk::NewMacro and itk::TypeMacro.

It is also mandatory to implement three methods in a new application:

  • DoInit()

  • DoUpdateParameters()

  • DoExecute()

DoInit()

This method is called once, when the application is instantiated. It should contain the following actions:

  • Set the name and the description of the application

  • Fill the documentation and give an example

  • Declare all the parameters

  • Define the documentation link: using for contrib application use SetDocLink("docLink"), for official application use SetOfficialDocLink().

DoUpdateParameters()

This method is called after every modification of a parameter value. With the command line launcher, it is called each time a parameter is loaded. With the Qt launcher, it is called each time a parameter field is modified. It can be used to maintain consistency and relationship between parameters (e.g. in ExtractROI: when changing the input image, maybe the ROI size has to be updated).

DoExecute()

This method contains the real action of the application. This is where the pipeline must be set up. The application framework provides different methods to get a value or an object associated to a parameter:

  • GetParameterInt(key) : get the integer value of a parameter

  • GetParameterFloat(key) : get the float value of a parameter

  • GetParameterString(key) : get the string value of a parameter

  • GetParameterImage(key) : get a pointer to an image object, read from the file name given in input

where key refers to parameter key, defined using AddParameter() method in DoInit() method.

Similar methods exist for binding a data object to an output parameter:

  • SetParameterOutputImage(key,data) : link the image object to the given output parameter

  • SetParameterOutputVectorData(key,data) : link the vector data object to the given output parameter

If possible, no filter update should be called inside this function. The update will be automatically called afterwards : for every image or vector data output, a writer is created and updated.

Parameters selection

In the new application framework, every input, output or parameter derive from otb::Wrapper::Parameter. The application engine supplies the following types of parameters:

  • ParameterType_Bool : parameter storing a boolean.

  • ParameterType_Int : parameter storing an integer.

  • ParameterType_Radius : parameter storing a radius.

  • ParameterType_Float : parameter storing a float.

  • ParameterType_String : parameter storing character string.

  • ParameterType_StringList : parameter storing a list of character string.

  • ParameterType_InputFilename : parameter storing an input file name.

  • ParameterType_InputFilenameList : parameter storing a list of input file names.

  • ParameterType_Directory : parameter storing a folder name.

  • ParameterType_Group : parameter storing children parameters.

  • ParameterType_Choice : parameter storing a list of choices (doesn’t support multi-choice). It also allows to create specific sub-parameters for each available choice.

  • ParameterType_ListView : parameter storing a list of choices (support multi-choice and single-choice).

  • ParameterType_InputImage : parameter storing an input image.

  • ParameterType_InputImageList : parameter storing a list of input image.

  • ParameterType_InputVectorData : parameter storing input vector data.

  • ParameterType_InputVectorDataList : parameter storing a list of input vector data.

  • ParameterType_OutputFilename : parameter storing an output file name.

  • ParameterType_OutputImage : parameter storing an output image.

  • ParameterType_OutputVectorData : parameter storing an output vector data.

  • ParameterType_RAM : parameter storing the maximum amount of RAM to be used.

  • ParameterType_Field : parameter storing a list of field from a given vector data (support multi-choice and single-choice).

  • ParameterType_Band : parameter storing a list of band from a given raster data (support multi-choice and single-choice).

Parameters description

Each created parameter has a unique key and several boolean flags to represent its state. These flags can be used to set a parameter optional or test if the user has modified the parameter value. The parameters are created in the DoInit() method, then the framework will set their value (either by parsing the command line or reading the graphical user interface). The DoExecute() method is called when all mandatory parameters have been given a value, which can be obtained with “Get” methods defined in otb::Wrapper::Application. Parameters are set mandatory (or not) using MandatoryOn(key) method (MandatoryOff(key)).

Some functions are specific to numeric parameters, such as SetMinimumParameterIntValue(key,value) or SetMaximumParameterFloatValue(key,value). By default, numeric parameters are treated as inputs. If your application outputs a number, you can use a numeric parameter and change its role by calling SetParameterRole(key,Role_Output).

The input types InputImage, InputImageList, InputVectorData and InputVectorDataList store the name of the files to load, but they also encapsulate the readers needed to produce the input data.

The output types OutputImage and OutputVectorData store the name of the files to write, but they also encapsulate the corresponding writers.

Composite application

The application framework has been extended to allow the implementation of composite applications : applications that use other applications. The concept is simple : you have two applications A and B that you want to chain in order to build a third application C. Rather than writing C by copying the code of A and B, you would like to re-use applications A and B. This plain example will be re-used in this section for explanations.

A dedicated class otb::Wrapper::CompositeApplication exists to create such applications. If you derive this class to implement application C, you will be able to create a composite application.

Creating internal applications

Like with standard applications, you have to write a DoInit() function. In this function, you should first clean any internal application with the function ClearApplications() (the DoInit() function is called twice in some cases). Then you can instantiate the internal applications that you want to use (for instance A and B). The function AddApplication() will do that, based on :

  • The application type (i.e. its official name, such as ExtractROI, BandMath, …)

  • An identifier : like with parameter keys, you have to specify an identifier to refer to this internal application. Use the same naming conventions as parameters.

  • A description : give a small description of the role of this internal application.

Using the function GetInternalApplication(), you can get a pointer to the internal application corresponding to a given identifier.

In the example given in introduction, we assume that :

  • An internal application of type A has been added with identifier a

  • An internal application of type B has been added with identifier b

Connecting parameters

Once you have internal applications, you may want to setup their parameters. There are typically 3 cases.

You may want to expose a parameter of an internal application as a parameter of your composite application. Let say you want to expose parameter io.in from application a into your composite application C with the key input. You can call the function :

ShareParameter(input,a.io.in)

As a result, the parameters input in application C and io.in in application a will point to the same object. Under the two parameter keys, there is a unique value. These two parameters can be considered as synchronized.

This leads to the second case : you may want to synchronize two parameters from internal applications. Let say you want to synchronize parameter field from application a with parameter fname from application b. You can call the function :

Connect(a.field,b.fname)

Note that the functions ShareParameter() and Connect() :

  • Use the same syntax to access internal parameters (“application identifier” dot “parameter key”).

  • Shall be used in the DoInit() function, after the internal applications have been added.

In this synchronization, the two parameters should have the same type, or have a similar interface, such as input and output filenames that are both accessed using GetParameterString() and SetParameterString().

This type of connection is a transition to the third case : you may want to connect the output of an internal application to the input of an other internal application. Here the difficulty is that the two parameters to connect probably have different types. Let say you want to connect parameter a.out to parameter b.in. The “Connect()” function may work in favorable cases (see previous paragraph), but for images, you have two options :

  • Explicitly copy the image pointer from the output image parameter in the input image parameter (with functions SetParameterInputImage() and GetParameterOutputImage()). It will connect the pipelines in applications A and B, to form an “in-memory” connection. This has to be done between the calls to DoExecute() of application A and B.

  • Use a temporary filename to store the output image a.out and read it with b.in. In this case, you have to manually call the writers of parameter a.out.

At the moment, the in-memory connection of vector data parameters is not supported.

Orchestration

In the DoUpdateParameters() of your composite application, you can call the same function on an internal application with the function UpdateInternalParameters(). This is needed only if your internal applications have a specific behaviour during parameter update.

In the DoExecute() of your composite application, you have to call ExecuteInternal() in order to launch each internal application. The order should be compatible with image parameter connexions. If you want to do “in-memory” connexions, you can do it between two calls to ExecuteInternal(), for instance :

ExecuteInternal("a");
GetInternalApplication("b")->SetParameterInputImage("in", GetInternalApplication("a")->GetParameterOutputImage("out"));
ExecuteInternal("b");

The application BundleToPerfectSensor is a simple example of composite applications. For a more complex example, you can check the application TrainImagesClassifier.

Compile your application

In order to compile your application you must call the macro OTB_CREATE_APPLICATION in the CMakelists.txt file. This macro generates the lib otbapp_XXX.so, in (OTB_BINARY_DIR/lib/otb/applications), where XXX refers to the class name.

Execute your application

There are different ways to launch applications :

CommandLine :

The command line option is invoked using otbApplicationLauncherCommandLine executable followed by the classname, the application dir and the application parameters.

Python :

A Python wrapper is also available.

Testing your application

It is possible to write application tests. They are quite similar to filters tests. The macro OTB_TEST_APPLICATION makes it easy to define a new test.

Application Example

See example ApplicationExample.cxx