---
license: mit
language:
- en
base_model:
- Ultralytics/YOLOv8
pipeline_tag: object-detection
tags:
- Ultralytics
- YOLOv8
---

# YOLOv8

This version of YOLOv8 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: 5.1

## Convert tools links:

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

- [The repo of AXera Platform](https://github.com/AXERA-TECH/ax-samples), which you can get the detial of guide

- [Pulsar2 Link, How to Convert ONNX to axmodel](https://pulsar2-docs.readthedocs.io/en/latest/pulsar2/introduction.html) 


## Support Platform

- AX650
  - [M4N-Dock(爱芯派Pro)](https://wiki.sipeed.com/hardware/zh/maixIV/m4ndock/m4ndock.html)
  - [M.2 Accelerator card](https://axcl-docs.readthedocs.io/zh-cn/latest/doc_guide_hardware.html)
- AX630C
  - [爱芯派2](https://axera-pi-2-docs-cn.readthedocs.io/zh-cn/latest/index.html)
  - [Module-LLM](https://docs.m5stack.com/zh_CN/module/Module-LLM)
  - [LLM630 Compute Kit](https://docs.m5stack.com/zh_CN/core/LLM630%20Compute%20Kit)
  
##Performance Statistics

#### AX650N

| Model | Latency(ms) npu1 | Latency(ms) npu3 |
| :--- | :---: | :---: |
| **yolo26n** | 4.112 | 1.410 |
| **yolo26s** | 10.173 | 3.549 |
| **yolo26m** | 26.824 | 9.389 |
| **yolo26l** | 51.532 | 17.858 |
| **yolo26x** | 88.379 | 28.892 |



#### AX630C

| Model | Latency(ms) npu1 | Latency(ms) npu2 |
| :--- | :---: | :---: |
| **yolo26n** | 14.489 | 8.308 |
| **yolo26s** | 31.349 | 23.601 |
| **yolo26m** | 80.694 | 57.040 |
| **yolo26l** | 143.709 | 100.611 |
| **yolo26x** | 244.731 | 158.686 |

#### AX615

| Model | Latency(ms) npu1 | Latency(ms) npu2 |
| :--- | :---: | :---: |
| **yolo26n** | 17.830 | 10.652 |
| **yolo26s** | 48.895 | 27.398 |
| **yolo26m** | 133.628 | 69.313 |

#### AX637
| Model | Latency(ms) npu1 |
| :--- | :---: |
| **yolo26n** | 4.714 |
| **yolo26s** | 11.945 |
| **yolo26m** | 28.219 |
| **yolo26l** | 53.790 |
| **yolo26x** | 84.507 |

## How to use

Download all files from this repository to the device

### Inference

Input image:
![](./bus.jpg)

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

```
root@ax650:~/ax650# python3 ax_infer.py --model-path yolov8m_640x640_npu3.axmodel --test-img bus.jpg
[INFO] Available providers:  ['AxEngineExecutionProvider', 'AXCLRTExecutionProvider']
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Chip type: ChipType.MC50
[INFO] VNPU type: VNPUType.DISABLED
[INFO] Engine version: 2.12.0s
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 6.0-dirty a498e20d-dirty
[YOLOv8-Det] [13:44:39.169] [DEBUG] Load model time = 483.73 ms
[YOLOv8-Det] [13:44:39.205] [DEBUG] Pre-process time = 7.49 ms
[YOLOv8-Det] [13:44:39.234] [DEBUG] Forward time = 28.59 ms
[YOLOv8-Det] [13:44:39.245] [DEBUG] Post-process time = 10.48 ms
[YOLOv8-Det] [13:44:39.249] [INFO] Draw Results (5 objects):
[YOLOv8-Det] [13:44:39.250] [INFO] (2, 229, 800, 741) -> bus: 0.97
[YOLOv8-Det] [13:44:39.253] [INFO] (51, 400, 247, 905) -> person: 0.93
[YOLOv8-Det] [13:44:39.254] [INFO] (668, 393, 808, 881) -> person: 0.91
[YOLOv8-Det] [13:44:39.255] [INFO] (221, 403, 343, 859) -> person: 0.89
[YOLOv8-Det] [13:44:39.255] [INFO] (0, 549, 78, 873) -> person: 0.79
[YOLOv8-Det] [13:44:39.281] [INFO] Saved to result_yolov8_det.jpg
--------------------------------------
```

Output image:
![](./result_yolov8_det.jpg)