Unit Testing & TensorRT Engine Improvements

[Ishaan-Datta]

Ultralytics-based Objects:

• Finish results, engine, model classes based off Ultralytics (/clean/ultralyics module)
• Make sure to use similar documentation style (docstrings) and add readme.md w/ UML for interface documentation/methods
• Finish other folders in /clean folder (make notes on any differences in our implementation and document design choices before deleting)

Testing:

• Finish the unit testing for the model output utilities using results object and inference utils
• Run unit test suites for both onnxruntime and TensorRT inference classes, report results (test cases are in /assets)

TensorRT Engine Extensions:

• In the [Google Drive](https://drive.google.com/drive/folders/1C16yonOKZXYdj6iSxf8vz8SOtjCeSww0?usp=drive_link

[aqil18]

What does the Results object do?

Shows you coordinates of where the bounding boxes are - produced by the model

Places where the Results class is used

ros2_ws/python/utils.py (Ultralytics) - This uses the ultralytics example
tools/conversion/model_exporter - This is not used in any actaul code, it's just an example
tools/testing/inference-results - This is where we are making our own results object
tools/utils/onxx_analysis - Creates its own results object for Onxx related analysis

Questions to ask

1. Differences between TensorRt, Onxx, Yolov8, Ultralytics
   ONXX is the format of model that allows us to run on different frameworks - can export with onnx or pytorch
   TensorRT is the Nvidia framework we run our models on
   Yolo is the model we build
   Ultralytics makes yolo possible

2. Which tools are we using and not using?
   a. Based on this which results object do we need? Do we need to create our own object?
   b. Clean up the code that uses the other tech imo

Next

Try using the Ultralytics module as is - creating our own objects is unnecessary if this is all pre-implemented
Start testing with the module