The Processor class allows users to implement processing or computation steps after DeepLabCut pose estimation. For example, a Processor can detect certain features of a pose and turn on an LED or an optogenetics laser, a Processor can implement a forward-prediction model that predicts animal's pose ~10-100 ms into the future to apply feedback with zero latency, or a Processor can do both.
The Processor is designed to be extremely flexible: it must only contain two methods: Processor.process, whose input and output is a pose as a numpy array, and Processor.save, which allows users to implement a method that saves any data the Processor acquires, such as the time that desired behavior occured or the times an LED or laser was turned on/off. The save method must be written by the user, so users can choose whether this data is saved as a text/csv, numpy, pickle, or pandas file to provide a few examples.
To write your own custom Processor, your class must inherit the base Processor class (see here):
from dlclive import Processor
class MyCustomProcessor(Processor):
...
To implement your processing steps, overwrite the Processor.process method:
def process(pose, **kwargs):
# my processing steps go here
return pose
For example Processor objects that communicate with Teensy microcontrollers to control an optogenetics laser, turn on an LED when upond detecting a mouse licking, or turn on an LED upon detecting a dog's rearing movement, see our example_teensy directory.