YOLO26-Seg
This version of YOLOv26-Seg has been converted to run on the Axera NPU using w8a16 quantization.
Compatible with Pulsar2 version: 4.2.
Convert tools links:
For those who are interested in model conversion, you can try to export axmodel through:
- The repo of AXera Platform, where you can get the detailed guide.
- Pulsar2 Link, How to Convert ONNX to axmodel
Support Platform
- AX650N/AX8850
- AX630C
- AX615
- AX637
Performance Statistics
AX650N(npu1)
| Model | FPS | CMM(MB) | Latency(ms) |
|---|---|---|---|
| yolo26n-seg | 201.86 | 5.49 | 4.954 |
| yolo26s-seg | 76.06 | 17.87 | 13.148 |
| yolo26m-seg | 23.67 | 39.79 | 42.256 |
| yolo26l-seg | 19.94 | 41.23 | 50.161 |
| yolo26x-seg | 8.83 | 85.31 | 113.295 |
AX650N(npu3)
| Model | FPS | CMM(MB) | Latency(ms) |
|---|---|---|---|
| yolo26n-seg | 507.10 | 3.58 | 1.972 |
| yolo26s-seg | 212.63 | 11.23 | 4.703 |
| yolo26m-seg | 70.05 | 31.00 | 14.275 |
| yolo26l-seg | 59.97 | 37.27 | 16.675 |
| yolo26x-seg | 27.25 | 77.87 | 36.701 |
AX630C
| Model | Latency(ms) npu1 | Latency(ms) npu2 |
|---|---|---|
| yolo26n-seg | 13.672 | 8.597 |
| yolo26s-seg | 29.745 | 22.234 |
AX615
| Model | Latency(ms) npu1 | Latency(ms) npu2 |
|---|---|---|
| yolo26n-seg | 24.046 | 14.558 |
| yolo26s-seg | 71.193 | 37.333 |
AX637
| Model | Latency(ms) |
|---|---|
| yolo26n-seg | 5.604 |
| yolo26s-seg | 15.271 |
How to use
Download all files from this repository to the device.
python env requirement
pyaxengine
https://github.com/AXERA-TECH/pyaxengine
wget https://github.com/AXERA-TECH/pyaxengine/releases/download/0.1.3.rc2/axengine-0.1.3-py3-none-any.whl
pip install axengine-0.1.3-py3-none-any.whl
others
Maybe None.
Inference with AX650 Host, such as M4N-Dock(爱芯派Pro)
Input image:
run
python ax_infer.py --model-path yolo26l-seg_npu3.axmodel --test-img bus.jpg
(ax_env) root@ax650:~/ax650# python ax_infer_seg.py --model-path yolo26l-seg_npu3.axmodel --test-img bus.jpg
[INFO] Available providers: ['AxEngineExecutionProvider']
[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: 5.1-patch1-dirty 9164b433-dirty
[YOLO26-Seg] [16:47:45.205] [DEBUG] Load model time = 821.55 ms
[YOLO26-Seg] [16:47:45.242] [DEBUG] Pre-process time = 9.03 ms
[YOLO26-Seg] [16:47:45.292] [DEBUG] Forward time = 49.07 ms
[YOLO26-Seg] [16:47:45.299] [DEBUG] Post-process time = 6.77 ms
[YOLO26-Seg] [16:47:45.300] [DEBUG] Proto shape: (32, 160, 160)
[YOLO26-Seg] [16:47:45.350] [INFO] Draw Results (6 objects):
[YOLO26-Seg] [16:47:45.350] [INFO] (10, 229, 803, 735) -> bus: 0.94
[YOLO26-Seg] [16:47:45.396] [INFO] (55, 396, 246, 905) -> person: 0.94
[YOLO26-Seg] [16:47:45.419] [INFO] (221, 406, 348, 855) -> person: 0.93
[YOLO26-Seg] [16:47:45.440] [INFO] (669, 392, 808, 878) -> person: 0.93
[YOLO26-Seg] [16:47:45.459] [INFO] (0, 549, 78, 874) -> person: 0.50
[YOLO26-Seg] [16:47:45.477] [INFO] (3, 532, 78, 1058) -> person: 0.45
[YOLO26-Seg] [16:47:45.521] [INFO] Saved to result_yolo26_seg.jpg
Output image:
- Downloads last month
- 56