Discover the power of real-time object detection on the STM32N6 !

In this video, we demonstrate how a YOLOv3-tiny model trained to detect colored marbles performs impressively on power-optimized embedded hardware. Watch as the STM32N6 detects a "large" number of objects in real-time, leveraging the efficiency of ST Cube.AI quantization to make edge AI faster and more energy-efficient.

The model was trained with Darknet from a limited dataset using a 416x416 input resolution. The video is recorded from the device USB UVC stream. Notice how the marble’s colored markers show some lag when too many have to be drawn for each frame… then, we ran out of marbles!