ramping up on roboflow supervision and trackers
i ran very room video through roboflow's DeepSORT tracker to understand what outputs look like and if this idea is feasible at all?
i observed that a detection object has bounding box coordinate, confidence score, (unique) object id. i think this is pretty useful information.
i accumulated all these detections in an array and one shotted this with vlm, just to see what happens. as expected it wasn't good and model outputted some gibberish.
3/n