A task is a unit of processing, applied to one or more inputs, and producing derived outputs.
- Some tasks produce no outputs, only messages in the console.
- Tasks often produce outputs independently for each corresponding input e.g. scaling each image to a different size.
- However, tasks can also produce outputs by aggregating many inputs e.g. a montage image of all inputs.
If you are familar with the map-reduce processing paradigm, these can be considered like a map step or a reduce step.
Several predefined tasks are bundled with the Anchor distribution - intended to be easily executed from the command line.
Users are free to add (or remove) tasks to this directory.
Any task in this directory can be referred in the Anchor command-line application by only its name, with neither a full path nor an
.xml extension. These statements both refer to the same task, and achieve the same effect:
anchor -i *.png -t features/hog anchor -i *.png -t /path/to/anchorDistribution/config/tasks/features/hog.xml
Understanding each task
- An overview of all the predefined tasks is found in Example Usage.
- Each task also includes a more detailed description as a comment in its BeanXML.
Building a new task
New tasks can be specified in BeanXML using classes that inherit from Task.