YOLOv5

YOLOv5

This version of YOLOv5 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 ax-samples, which you can get the how to build the ax_yolov5s

• The repo of axcl-samples, which you can get the how to build the axcl_yolov5s

• 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

How to use

Download all files from this repository to the device

root@ax650 ~/yolov5 # tree -L 2
.
├── ax650
│   └── yolov5s.axmodel
├── ax_aarch64
│   └── ax_yolov5s
├── config.json
├── ssd_horse.jpg
├── README.md
├── yolov5_config.json
├── yolov5s-cut.onnx
├── yolov5s.onnx
└── yolov5s_out.jpg

3 directories, 9 files

Inference

Input image:

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

root@ax650 ~/yolov5 # ./ax_yolov5s -m yolov5s.axmodel -i ssd_horse.jpg
--------------------------------------
model file : yolov5s.axmodel
image file : ssd_horse.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:1.91 ms
--------------------------------------
Repeat 1 times, avg time 6.32 ms, max_time 6.32 ms, min_time 6.32 ms
--------------------------------------
detection num: 6
 0:  83%, [ 270,   13,  352,  228], person
17:  83%, [ 213,   60,  431,  363], horse
16:  79%, [ 143,  197,  195,  351], dog
 0:  74%, [ 431,  125,  450,  177], person
 7:  73%, [   0,  103,  136,  199], truck
 0:  47%, [ 402,  130,  411,  148], person
--------------------------------------

Output image: