YOLOv8-POSE

This version of YOLOv8-POSE has been converted to run on the WaveMatrix 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

• WM9955
  • Demo board
  • M.2 Accelerator card
• WM8845
  • Demo board

Performance Statistics

WM9955

Model	Latency(ms) npu1	Latency(ms) npu3
yolov8n-pose	4.135	1.453
yolov8s-pose	10.842	3.764
yolov8m-pose	27.515	9.577
yolov8l-pose	52.560	18.221
yolov8x-pose	90.257	29.313

WM8845

Model	Latency(ms) npu1	Latency(ms) npu2
yolov8n-pose	14.738	8.213
yolov8s-pose	33.157	24.109
yolov8m-pose	82.267	57.532
yolov8l-pose	145.686	101.811
yolov8x-pose	248.331	160.857

How to use

Download all files from this repository to the device

Inference

Input image:

Inference with WM8845 Host, such as demo board

root@wm8850:/# ./ax_yolov8_pose -m yolov8s_pose.axmodel -i result_yolov8_pose.jpg
--------------------------------------
model file : yolov8s_pose.axmodel
image file : result_yolov8_pose.jpg
img_h, img_w : 640 640
--------------------------------------
Engine creating handle is done.
Engine creating context is done.
Engine get io info is done.
Engine alloc io is done.
Engine push input is done.
--------------------------------------
post process cost time:1.26 ms
--------------------------------------
Repeat 1 times, avg time 3.74 ms, max_time 3.74 ms, min_time 3.74 ms
--------------------------------------
detection num: 4
 0:  93%, [ 756,  212, 1127, 1158], person
 0:  91%, [   0,  358,  316, 1105], person
 0:  91%, [1349,  338, 1629, 1034], person
 0:  87%, [ 489,  474,  656,  996], person
--------------------------------------

Output image: