Skip to main content

YOLOv8 Pose

This document describes how to run YOLOv8 Pose on NPU.

info

Refer to Model Zoo Download for the example.

YOLOv8 Pose Example Directory Structure:

$ tree ./
./
├── CMakeLists.txt
├── convert_model
│   ├── config_yml.py
│   ├── convert_model_env.sh
│   ├── python
│   │   ├── onnx_extract.py
│   │   └── yolov8s-pose_640.txt
│   └── yolov8s-pose_9.txt
├── figures
│   ├── diff_img.png
│   └── out_yolov8_pose_pcq.png
├── main.cpp
├── model
│   └── COCO_train2014_000000500390.jpg
├── model_config.h
├── README.md
├── yolov8_pose_9_post.cpp
└── yolov8_pose_9_pre.cpp

Model Conversion​

Configure Virtual Environment​

X86 Linux PC
python -m venv .venv && source .venv/bin/activate
pip install ultralytics==8.1.0 onnxsim

Export ONNX Model​

X86 Linux PC
cd convert_model/python/
yolo export model=yolov8s-pose.pt format=onnx dynamic=True opset=11

Fixed Shape​

X86 Linux PC
python3 -m onnxsim yolov8s-pose.onnx yolov8s-pose_640.onnx --input-shape=1,3,640,640

Prune Model​

X86 Linux PC
python3 onnx_extract.py
cd ..
X86 Linux PC
./convert_model_env.sh

Model Import/Quantization/Conversion​

You need to enter the container development environment first. Refer to the Create Container section in Model Zoo Download.

info

Different platforms use corresponding Docker images:

  • A733: ubuntu-npu:v2.0.10.1
  • T527: ubuntu-npu:v1.8.11
X86 Linux PC
docker exec -it model-zoo /bin/bash

After entering the container, navigate to the corresponding directory and run the script.

X86 Linux PC
cd /workspace/examples/yolov8_pose/convert_model/
X86 Linux PC
./pegasus_import.sh yolov8s-pose_9
./pegasus_quantize.sh yolov8s-pose_9 uint8 12
X86 Linux PC
./pegasus_export_ovx_nbg.sh yolov8s-pose_9 uint8 a733

The exported model files are stored in the ../model directory.

Compile Example​

Now you can compile the example. First exit the container, then execute the following command to compile the example.

First, you need to configure third-party libraries and cross-compilation toolchain.

info

You can skip this step if you have already configured third-party libraries and cross-compilation toolchain in other examples.

X86 Linux PC
cd ../../../3rdparty/opencv/
unzip opencv-4.9.0-aarch64-linux-sunxi-glibc.zip
cd ../../0-toolchains/

You need to manually download via this link first, then place it in 0-toolchains/ before executing the following command:

X86 Linux PC
tar -xvf gcc-arm-10.2-2020.11-x86_64-aarch64-none-linux-gnu.tar.xz
X86 Linux PC
cd ../examples/yolov8_pose/
X86 Linux PC
../build_linux.sh -t a733 -s debian11

Model Deployment​

After compilation, the example will be installed in the install directory. You can use scp to transfer it to the board.

Configure NPU Driver​

info

You can skip this step if you have already configured NPU driver in other examples.

Transfer the driver library to the board's lib directory via scp.

  • A733 corresponds to the common/lib_linux_aarch64/A733 directory
  • T527 corresponds to the common/lib_linux_aarch64/T527 directory

Then execute the following command to export to environment variables.

Radxa SBC
echo 'export LD_LIBRARY_PATH=$HOME/lib:$LD_LIBRARY_PATH' >> ~/.bashrc

Run Example​

After configuring the driver, you can run the example.

tip

For T527 platform, you need to first enable NPU by referring to the A5E's "Enable NPU on Board" documentation, then use the following command to grant the current user permission to use /dev/vipcore.

Radxa SBC
sudo chmod 777 /dev/vipcore
Radxa SBC
cd yolov8_pose_demo_linux_a733/
Radxa SBC
chmod +x ./yolov8_pose_demo_a733
./yolov8_pose_demo_a733 -nb model/yolov8s-pose_9_uint8_a733.nb -i model/COCO_train2014_000000500390.jpg

The running result is as follows:

$ ./yolov8_pose_demo_a733 -nb model/yolov8s-pose_9_uint8_a733.nb -i model/COCO_train2014_000000500390.jpg
model_file=model/yolov8s-pose_9_uint8_a733.nb, input=model/COCO_train2014_000000500390.jpg, loop_count=1, malloc_mbyte=10
VIPLite driver software version 2.0.3.2-AW-2024-08-30
input  0 dim 3 640 640 1, data_format=2, quant_format=0, name=input/output[0], none-quant
output 0 dim 80 80 64 1, data_format=0, name=uid_17_out_0b_uid_1_out_0, none-quant
output 1 dim 80 80 1 1, data_format=0, name=uid_16_out_0b_uid_1_out_0, none-quant
output 2 dim 80 80 51 1, data_format=0, name=uid_15_out_0b_uid_1_out_0, none-quant
output 3 dim 40 40 64 1, data_format=0, name=uid_14_out_0b_uid_1_out_0, none-quant
output 4 dim 40 40 1 1, data_format=0, name=uid_13_out_0b_uid_1_out_0, none-quant
output 5 dim 40 40 51 1, data_format=0, name=uid_12_out_0b_uid_1_out_0, none-quant
output 6 dim 20 20 64 1, data_format=0, name=uid_11_out_0b_uid_1_out_0, none-quant
output 7 dim 20 20 1 1, data_format=0, name=uid_10_out_0b_uid_1_out_0, none-quant
output 8 dim 20 20 51 1, data_format=0, name=uid_9_out_0ub_uid_1_out_0, none-quant
nbg name=model/yolov8s-pose_9_uint8_a733.nb, size: 7768344.
create network 0: 18985 us.
prepare network: 5711 us.
buffer ptr: 0x28344380, buffer size: 1228800
network: 0, loop count: 1
run time for this network 0: 32958 us.
output 0, ptr 0x28470480, size 409600.
output 1, ptr 0x28600500, size 6400.
output 2, ptr 0x28606980, size 326400.
output 3, ptr 0x28745640, size 102400.
output 4, ptr 0x287a96c0, size 1600.
output 5, ptr 0x287ab040, size 81600.
output 6, ptr 0x287fabc0, size 25600.
output 7, ptr 0x28813c80, size 400.
output 8, ptr 0x28814340, size 20400.
post process time : 4 ms
detection num: 3
 0:  93%, [ 373,    1,  587,  346], person
411.58 41.32 = 0.98922
419.64 35.78 = 0.98396
416.36 37.19 = 0.76423
440.57 37.33 = 0.97060
422.52 38.08 = 0.12822
450.85 69.58 = 0.99924
422.50 75.59 = 0.99804
473.26 121.09 = 0.99354
405.71 108.13 = 0.95213
449.46 97.35 = 0.98640
389.93 81.26 = 0.92587
461.08 161.27 = 0.99969
461.53 162.40 = 0.99945
405.04 226.47 = 0.99954
489.77 240.76 = 0.99892
415.75 320.01 = 0.99481
555.85 276.33 = 0.99307
 0:  93%, [  86,   28,  288,  390], person
155.68 76.87 = 0.99271
162.21 68.34 = 0.98739
145.10 65.45 = 0.95864
175.03 64.92 = 0.91619
141.19 64.45 = 0.68796
199.98 93.83 = 0.99730
160.28 98.94 = 0.99395
214.27 138.31 = 0.99138
164.10 156.53 = 0.98026
175.57 174.47 = 0.98414
136.65 193.82 = 0.97464
216.19 199.03 = 0.99952
180.07 198.95 = 0.99935
240.79 270.84 = 0.99790
150.18 279.74 = 0.99727
293.96 281.26 = 0.98766
128.72 359.77 = 0.98534
 0:  91%, [ 227,   36,  398,  405], person
281.36 106.56 = 0.99230
287.73 97.59 = 0.98680
279.19 104.41 = 0.88281
308.31 83.30 = 0.95046
275.74 96.24 = 0.19450
328.64 102.08 = 0.99900
275.67 126.89 = 0.99741
373.27 126.41 = 0.99145
278.76 161.70 = 0.95360
382.76 163.57 = 0.98179
249.65 205.81 = 0.91721
332.66 214.56 = 0.99969
309.45 218.78 = 0.99948
293.53 304.20 = 0.99923
310.05 306.35 = 0.99817
279.65 380.53 = 0.99397
363.50 304.91 = 0.99248
destroy npu finished.
~NpuUint.

This performance data only calculates the time consumption of model inference. Unless otherwise specified, it does not include the time consumption of pre-processing and post-processing.

SoCNPUModelInput ResolutionNetwork Creation TimeNetwork Preparation TimeSingle Frame Inference TimePost-processing TimeTotal TimeFrame Rate
Allwinner A733Vivante VIP9000yolov8s-pose640×64019.0 ms5.7 ms33.0 ms4.0 ms61.7 ms30.3 FPS

    You need to be logged into GitHub to post a comment. If you are already logged in, please ignore this message.

    Radxa-docs © 2026 by Radxa Computer (Shenzhen) Co.,Ltd. is licensed under CC BY 4.0