Depends on specific cases, I have on good authority of how in few "bleeding edge" ones they essentially repacked/wrapped YOLOv3. Purpose was specifically tracking in adversarial conditions (smoke, including smokescreen, obstacles, etc)
For realtime on the edge the YOLO series is pretty good, I don't think anyone would disagree. Most of the really advanced stuff like Vision Language models all require a lot more compute and power budget.
p_l|3 days ago
Onavo|3 days ago
nextaccountic|2 days ago
nextaccountic|2 days ago