YOLOv8-Seg

This version of YOLOv8-Seg 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: 3.4

Convert tools links:

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

• The repo of AXera Platform, which you can get the detial of guide

• Pulsar2 Link, How to Convert ONNX to axmodel

Support Platform

• AX650
  • M4N-Dock(爱芯派Pro)
  • M.2 Accelerator card
• AX630C
  • 爱芯派2
  • Module-LLM
  • LLM630 Compute Kit

Chips	yolov8s-seg
AX650	4.6 ms
AX630C	TBD ms

How to use

Download all files from this repository to the device

root@ax650:~/YOLOv8-Seg# tree
.
|-- ax650
|   `-- yolov8s-seg.axmodel
|-- ax_yolov8_seg
|-- football.jpg
`-- yolov8_seg_out.jpg

Inference

Input image:

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

root@ax650:~/samples/AXERA-TECH/YOLOv8-Seg# ./ax_yolov8_seg -m ax650/yolov8s_seg.axmodel -i football.jpg
--------------------------------------
model file : ax650/yolov8s_seg.axmodel
image file : football.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.
--------------------------------------

input size: 1
    name:   images [UINT8] [BGR]
        1 x 640 x 640 x 3


output size: 7
    name: /model.22/Concat_1_output_0 [FLOAT32]
        1 x 80 x 80 x 144

    name: /model.22/Concat_2_output_0 [FLOAT32]
        1 x 40 x 40 x 144

    name: /model.22/Concat_3_output_0 [FLOAT32]
        1 x 20 x 20 x 144

    name: /model.22/cv4.0/cv4.0.2/Conv_output_0 [FLOAT32]
        1 x 80 x 80 x 32

    name: /model.22/cv4.1/cv4.1.2/Conv_output_0 [FLOAT32]
        1 x 40 x 40 x 32

    name: /model.22/cv4.2/cv4.2.2/Conv_output_0 [FLOAT32]
        1 x 20 x 20 x 32

    name:  output1 [FLOAT32]
        1 x 32 x 160 x 160

post process cost time:16.21 ms
--------------------------------------
Repeat 1 times, avg time 4.69 ms, max_time 4.69 ms, min_time 4.69 ms
--------------------------------------
detection num: 8
 0:  92%, [1354,  340, 1629, 1035], person
 0:  91%, [   5,  359,  314, 1108], person
 0:  91%, [ 759,  220, 1121, 1153], person
 0:  88%, [ 490,  476,  661,  999], person
32:  73%, [1233,  877, 1286,  923], sports ball
32:  63%, [ 772,  888,  828,  937], sports ball
32:  63%, [ 450,  882,  475,  902], sports ball
 0:  55%, [1838,  690, 1907,  811], person
--------------------------------------

Output image: