YOLOv8-POSE

This version of YOLOv8-POSE has been converted to run on the Axera NPU using w8a16 quantization.

This model has been optimized with the following LoRA:

Compatible with Pulsar2 version: 5.0

Convert tools links:

For those who are interested in model conversion, you can try to export axmodel through

• The repo of ax-samples, which you can get the how to build the ax_yolov8_pose

• The repo of axcl-samples, which you can get the how to build the axcl_yolov8_pose

• Pulsar2 Link, How to Convert ONNX to axmodel

Support Platform

• AX650N/AX8850
  • M4N-Dock(爱芯派Pro)
  • M.2 Accelerator card
• AX630C
  • 爱芯派2
  • Module-LLM
  • LLM630 Compute Kit
• AX615
• AX637

Performance Statistics

AX650N

AX630C

AX615

AX637

How to use

Download all files from this repository to the device

Inference

Input image:

Inference with AX650 Host, such as M4N-Dock(爱芯派Pro)

(base) root@ax650:~/ax650pose# python3 ax_infer.py --model-path yolov8s-pose_640x640_npu3.axmodel --test-img bus.jpg
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Chip type: ChipType.MC50
[INFO] VNPU type: VNPUType.DISABLED
[INFO] Engine version: 2.12.0s
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 6.0 a498e20d
[YOLOv8-Pose] [15:35:43.230] [DEBUG] Load model time = 576.32 ms
[YOLOv8-Pose] [15:35:43.301] [DEBUG] Pre-process time = 21.58 ms
[YOLOv8-Pose] [15:35:43.328] [DEBUG] Forward time = 26.25 ms
[YOLOv8-Pose] [15:35:43.335] [DEBUG] Post-process time = 6.46 ms
[YOLOv8-Pose] [15:35:43.341] [INFO] Draw Results (4 persons):
[YOLOv8-Pose] [15:35:43.342] [INFO] (222, 404, 344, 859) -> person: 0.89
[YOLOv8-Pose] [15:35:43.347] [INFO] (49, 397, 244, 902) -> person: 0.88
[YOLOv8-Pose] [15:35:43.348] [INFO] (669, 392, 808, 875) -> person: 0.87
[YOLOv8-Pose] [15:35:43.349] [INFO] (0, 413, 78, 935) -> person: 0.57
[YOLOv8-Pose] [15:35:43.377] [INFO] Saved to result_yolov8_pose.jpg

Output image: