TODO: Distill models using recorded training data

[burningion]

Use captured bicycle video recordings to distill our ViT and SAM models used for inference.

[burningion]

Looking at the gst-nvtracker page, I found a link to TrafficCamNet, which should give us decent results right out of the box.

Going to try downloading the model and creating a TensorRT to see what kind of performance I can get out of the Jetson Orin Nano.

[burningion]

Added the TensorRT model to models/, created TensorRT model by running the repo container, then running:

$ /usr/src/tensorrt/bin/trtexec --onnx=resnet18_trafficcamnet_pruned.onnx --saveEngine=resnet18_trafficcamera_pruned.engine

I did this in the directory I copied the source model to.