Implement common post_processing

[MarcA711]

This PR implements a structure to store post_processing in a common place for all detectors to avoid duplicate code and make it easier to support new models.

Steps to invoke post_processing from a detector:

class Rknn(DetectionApi):
    def __init__(self, config: RknnDetectorConfig):
        super().__init__(config)
        ...
 
    def detect_raw(self, tensor_input):
        output = inference(tensor_input)
        return self.post_process(output)

In case you want to implement your own post-processing for a model (for example to use hardware acceleration using opencl), reimplement the method like this:

def yolonas(self, output):
    if gpu_found:
        # implement hw post_processing ...
        return results
    else:
        return super().yolonas(output)

However, use this only if your implementation is significantly faster than the default implementation.

Steps to add a new model:

• add entry to ModelTypeEnum in frigate/detector/detector_config.py
• add model identifier to docs/docs/reference.md
• implement a new method in frigate/detectors/detection_api.py
  • try to implement post-processing using numpy, as all platforms use this
• call the method from post_processing in frigate/detectors/detection_api.py

[netlify]

✅ Deploy Preview for frigate-docs canceled.

Name	Link
🔨 Latest commit	aded9c9
🔍 Latest deploy log	https://app.netlify.com/sites/frigate-docs/deploys/66478707dfbe690008b8a4c0