Model Deployment

Owned by Shrinjay Mukherjee, created
Last updated: 2021-04-27
1 min read

Objective

Deploy both YOLO models into the targetAcquisition class. Allow for access from predict method and allow for switching between models.

Assumptions

  • Both YOLO models are prebuilt.

  • Training functionality is encapsulated within prebuilt model.

  • targetAcquisition acts as a unified front-end to the prebuilt models.

Requirements

Requirement

Implementation

Requirement

Implementation

1

Box Model Presence

Model will be contained in its own subfolder in targetAcquisition, structure unchanged. Already done.

2

Box Model Run

Already implemented in Taxi class

3

Pylon Model Presence

Download yolov5 model used in notebook and replace yolov2_assets with yolov5: GitHub - zldrobit/yolov5: YOLOv5 in PyTorch > ONNX > CoreML > iOS Re-use box model

4

Pylon Model Run

We need to retool the detect function in detect.py to get source from an image passed in, look into how the LoadImages function runs and retool it. Call detect from detect.py with image from pipeline passed in. Reconfigure detect function to return det rather than dump into folder. Re-use box model code

5

Multiprocessing

Set up pipeline to move results of det out of target acquisition. Write worker function for target acquisition once updated.