Skip to main content

ai-object-detection-realtime

:::warning [UNTESTED] — v0.5 paper-correct skeleton Builds clean on native_sim/native/64 + the AEN fallback target. Real DEEPX DX-M1 dispatch + DXNN-compiled YOLOv8-tiny + V2N MIPI camera capture land in v0.6 alongside V2N-M1 HiL bring-up. :::

Realtime YOLOv8-tiny object detection on a DEEPX DX-M1 NPU. Camera frames flow through <alp/camera.h>, the model runs on the on-module NPU via <alp/inference.h>, the bounding-box overlay + live FPS counter render through <alp/display.h>.

Source: examples/ai-object-detection-realtime/.

Pipeline​

OV5640 → <alp/camera.h> → <alp/inference.h> → <alp/display.h>
(MIPI)    RGB565            YOLOv8-tiny.dxnn    bbox overlay + FPS
                            on DEEPX DX-M1

board.yaml​

som:
  sku: E1M-V2M101

preset: e1m-x-evk

cores:
  a55_cluster:
    app: ./linux
    image: alp-image-edge
    peripherals: [i2c, spi]
    libraries: [lvgl]
    inference: { tensor_arena_kb: 4096 }   # app opens with .backend = ALP_INFERENCE_BACKEND_DEEPX_DXM1

chips:
  - ov5640
  - deepx_dxm1

diagnostics:
  log_level: info

See also​

Questions about this page? Discuss in Community Forum