Add bytrack to object detection model

[caleb-vicente]

In theory this type of models can improve the performance of bounding boxes in videos: https://github.com/ifzhang/ByteTrack

optical flow: https://www.kaggle.com/code/daigohirooka/optical-flow-estimation-using-raft

[caleb-vicente]

the output of raft looks like it is (Δx, Δy). where (0,0) is bottom, left corner:

https://medium.com/axinc-ai/raft-a-machine-learning-model-for-estimating-optical-flow-6ab6d077e178