YOLOv8-pose Example

Usage

Make sure you have downloaded the data files first for the examples. You only need to do this once for all examples.

cd example/
git clone --depth=1 https://github.com/swdee/go-rknnlite-data.git data

Run the YOLOv8-pose example on rk3588 or replace with your Platform model.

cd example/yolov8-pose
go run yolov8-pose.go -p rk3588

This will result in the output of:

Driver Version: 0.9.6, API Version: 2.3.0 (c949ad889d@2024-11-07T11:35:33)
Model Input Number: 1, Ouput Number: 4
Input tensors:
  index=0, name=images, n_dims=4, dims=[1, 640, 640, 3], n_elems=1228800, size=1228800, fmt=NHWC, type=INT8, qnt_type=AFFINE, zp=-128, scale=0.003922
Output tensors:
  index=0, name=/model.22/Concat_1_output_0, n_dims=4, dims=[1, 65, 80, 80], n_elems=416000, size=416000, fmt=NCHW, type=INT8, qnt_type=AFFINE, zp=31, scale=0.141366
  index=1, name=/model.22/Concat_2_output_0, n_dims=4, dims=[1, 65, 40, 40], n_elems=104000, size=104000, fmt=NCHW, type=INT8, qnt_type=AFFINE, zp=67, scale=0.235531
  index=2, name=/model.22/Concat_3_output_0, n_dims=4, dims=[1, 65, 20, 20], n_elems=26000, size=26000, fmt=NCHW, type=INT8, qnt_type=AFFINE, zp=50, scale=0.159995
  index=3, name=/model.22/Concat_6_output_0, n_dims=4, dims=[1, 17, 3, 8400], n_elems=428400, size=856800, fmt=NCHW, type=FP16, qnt_type=AFFINE, zp=0, scale=1.000000
person @ (49 104 168 526) 0.903778
person @ (348 94 486 516) 0.903778
person @ (472 116 579 525) 0.903778
person @ (251 143 363 520) 0.872132
person @ (678 112 779 526) 0.872132
person @ (161 109 266 521) 0.872132
person @ (572 103 686 528) 0.853203
Model first run speed: inference=37.624844ms, post processing=465.199µs, rendering=1.815006ms, total time=39.905049ms
Saved object detection result to ../data/people-yolov8-pose-out.jpg
Benchmark time=2.860705252s, count=100, average total time=28.607052ms
done

The saved JPG image with pose estimation markers.

See the help for command line parameters.

$ go run yolov8-pose.go --help

Usage of /tmp/go-build2381700544/b001/exe/yolov8-pose:
  -i string
        Image file to run object detection on (default "../data/people-poses.jpg")
  -l string
        Text file containing model labels (default "../data/yolov8_pose_labels_list.txt")
  -m string
        RKNN compiled YOLO model file (default "../data/models/rk3588/yolov8n-pose-rk3588.rknn")
  -o string
        The output JPG file with pose detection markers (default "../data/people-yolov8-pose-out.jpg")
  -p string
        Rockchip CPU Model number [rk3562|rk3566|rk3568|rk3576|rk3582|rk3582|rk3588] (default "rk3588")

Docker

To run the YOLOv8-pose example using the prebuilt docker image, make sure the data files have been downloaded first, then run.

# from project root directory

docker run --rm \
  --device /dev/dri:/dev/dri \
  -v "$(pwd):/go/src/app" \
  -v "$(pwd)/example/data:/go/src/data" \
  -v "/usr/include/rknn_api.h:/usr/include/rknn_api.h" \
  -v "/usr/lib/librknnrt.so:/usr/lib/librknnrt.so" \
  -w /go/src/app \
  swdee/go-rknnlite:latest \
  go run ./example/yolov8-pose/yolov8-pose.go -p rk3588

Benchmarks

The following table shows a comparison of the benchmark results across the three distinct platforms.

Platform	Execution Time	Average Inference Time Per Image
rk3588	2.86s	28.60ms
rk3576	3.64s	36.49ms
rk3566	5.48s	54.84ms

Note that these examples are only using a single NPU core to run inference on. The results would be different when running a Pool of models using all NPU cores available. Secondly the Rock 4D (rk3576) has DDR5 memory versus the Rock 5B (rk3588) with slower DDR4 memory.

Background

This YOLOv8-pose example is a Go conversion of the C API example.