YOLOv7

YOLOv7 on reComputer R Series (CM5 + Hailo-10H)

YOLOv7 / YOLOv7-tiny / YOLOv7e6 / YOLOv7x perform COCO 80-class object detection on Hailo-10H through HailoRT. The compiled HEF runs NMS on-chip (HPP, meta_arch=yolov5, at most 80 boxes per class), so the application only parses the compact NMS-by-class buffer HailoRT 5.1.1 returns.

This page targets reComputer R Series (CM5 + Hailo-10H) with a PCIe Hailo-10H accelerator.

Hardware and Host Setup

Install the Hailo-10H packages on the host and confirm that the driver and device are ready before starting the container:

sudo apt update
sudo apt install hailo-h10-all -y
sudo reboot

# After reboot
hailortcli --version
hailortcli fw-control identify
ls -l /dev/hailo0

The expected HailoRT version for these images is 5.1.1.

Run With Demo Video

YOLOv7:

sudo docker run --rm \
  --name cm5-hailo10h-yolov7 \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  -v /usr/lib/libhailort.so.5.1.1:/usr/lib/libhailort.so.5.1.1:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov7:latest \
  python web_detection.py --model_path model/yolov7.hef --video_path video/test.mp4

YOLOv7-tiny:

sudo docker run --rm \
  --name cm5-hailo10h-yolov7_tiny \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  -v /usr/lib/libhailort.so.5.1.1:/usr/lib/libhailort.so.5.1.1:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov7_tiny:latest \
  python web_detection.py --model_path model/yolov7_tiny.hef --video_path video/test.mp4

YOLOv7e6:

sudo docker run --rm \
  --name cm5-hailo10h-yolov7e6 \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  -v /usr/lib/libhailort.so.5.1.1:/usr/lib/libhailort.so.5.1.1:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov7e6:latest \
  python web_detection.py --model_path model/yolov7e6.hef --video_path video/test.mp4

YOLOv7x:

sudo docker run --rm \
  --name cm5-hailo10h-yolov7x \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  -v /usr/lib/libhailort.so.5.1.1:/usr/lib/libhailort.so.5.1.1:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov7x:latest \
  python web_detection.py --model_path model/yolov7x.hef --video_path video/test.mp4

Open http://<Board_IP>:8000 to view the web preview.

USB Camera Mode

sudo docker run --rm \
  --name cm5-hailo10h-yolov7 \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  --device /dev/video0:/dev/video0 \
  -v /usr/lib/libhailort.so.5.1.1:/usr/lib/libhailort.so.5.1.1:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov7:latest \
  python web_detection.py --model_path model/yolov7.hef --camera_id 0

Swap yolov7 for yolov7_tiny, yolov7e6 or yolov7x to use another variant.

HEF Files

All four variants are Hailo Model Zoo v5.4.0 Hailo-10H builds. mAP and FPS are the Model Zoo's published Hailo-10H numbers.

Source: https://hailo-model-zoo.s3.eu-west-2.amazonaws.com/ModelZoo/Compiled/v5.4.0/hailo10h/yolov7.hef (yolov7_tiny.hef, yolov7e6.hef, yolov7x.hef in the same directory).

REST API

Each variant has its own endpoint:

curl -X POST "http://<Board_IP>:8000/api/models/yolov7/predict" \
  -F "file=@bus.jpg" -F "conf=0.25"

The response contains predictions entries with class, confidence and box fields.

Implementation Notes

• The ALLS recipe is change_output_activation(sigmoid) plus nms_postprocess(<model>_nms_config.json, meta_arch=yolov5, engine=cpu); the NMS config uses nms_scores_th 0.2, nms_iou_th 0.6 and max_proposals_per_class 80, so the host runs no box decode and no NMS.
• HPP output on HailoRT 5.1.1 is the flat compact NMS-by-class buffer (per class: detection count followed by that many [ymin, xmin, ymax, xmax, score] rows, all normalized to [0, 1]). Fixed-stride readers misalign every class after the first; the parser stops at the first implausible header and also keeps handlers for the ragged NMS-by-score and dense (1,C,5,D) / (1,C,D,5) layouts.
• yolov7e6 uses a 1280x1280x3 input; the other three use 640x640x3. The application reads the size from the HEF at startup.
• Pre-processing letterboxes to the HEF input size with gray (114) padding, converts BGR to RGB and feeds raw uint8; the /255 normalization is compiled into the HEF.
• Class IDs 0..79 index the standard COCO class list directly.
• First-inference diagnostics are disabled in production (_OUTPUT_LOGGED = True); set it to False in web_detection.py to dump the raw output type, shape, value range and per-class rows when validating a new HEF.

Development Notes

• Source modules: src/hailo10h_yolov7/, src/hailo10h_yolov7_tiny/, src/hailo10h_yolov7e6/, src/hailo10h_yolov7x/ in the reComputer-Hailo10H-CV repository
• Containers: ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov7{,_tiny,e6,x}:latest
• The Hailo Model Zoo also publishes Hailo-8 builds of all four variants (v2.19.0 / v2.20.0); only the Hailo-10H modules are published here so far.
• Model licence: GPL-3.0 (upstream WongKinYiu/yolov7).