STM32N6570-DK object detection demo flags people only

The write-up follows an earlier installment introducing ST's STM32 model zoo, and this time the author moves from setup to actually running a sample. The board carries an STM32N657X0 microcontroller with an Arm Cortex-M55 core, a 600-Gops ST Neural-ART Accelerator, a Neo-Chrom 2.5D GPU and 4.2 MB of contiguous SRAM, and the bundled camera is based on Sony's IMX335 (the schematic lists AS-BC538A3M6-63 from HONGJIA, with possible future changes).

A notable practical wrinkle is that the STM32N6570 lacks internal flash memory, so flashing is done in development mode via SW1, with the external MX66UW1G45G memory selected as the loader at start address 0x70000000. The single provided hex file bundles bootloader, AI network and application segments at separate addresses (0x70000000, 0x70100000 and 0x70380000 respectively), letting a user rewrite only the network file when iterating on a model. The bootloader is shared across other AI applications in the same folder, so it need not be replaced when switching demos.

The detection result itself is modest in scope: the configuration file defines only "person" as a class, and the hands-on check matched that expectation, with whole-body subjects recognized at nearly 100% but partial views proving harder. The author used a NUCLEO-N657X0-Q board to capture images and plans to try other AI model data next.