YOLO26

This version of YOLOv26 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: 4.2.

Convert tools links:

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

Support Platform

Performance Statistics

AX650N (NPU3 Mode)

Model FPS CMM(MB) Latency(ms)
yolo26n 726 3.26 1.378
yolo26s 316 10.2 3.166
yolo26m 116 27.6 8.644
yolo26l 90 33.88 11.174
yolo26x 41.0 70.4 20.405

AX630C

Model Latency(ms)
yolo26n_npu1.axmodel 10.706
yolo26s_npu1.axmodel 23.188
yolo26n_npu2.axmodel 6.309
yolo26s_npu2.axmodel 16.347

AX637

Model Latency(ms)
yolo26n 4.059
yolo26s 10.933

AX615

Model Latency(ms)
yolo26n_npu1.axmodel 26.995
yolo26s_npu1.axmodel 55.200
yolo26n_npu2.axmodel 10.599
yolo26s_npu2.axmodel 18.600

How to use

Download all files from this repository to the device

root@ax650:~/YOLO26# tree .
|-- ax650
|   |-- yolo26n.axmodel
|   |-- yolo26s.axmodel
|   |-- yolo26m.axmodel
|   |-- yolo26l.axmodel
|   `-- yolo26x.axmodel
|-- ax630c
|   |-- yolo26n_npu1.axmodel
|   |-- yolo26s_npu1.axmodel
|   |-- yolo26n_npu2.axmodel
|   |-- yolo26s_npu2.axmodel
|-- ax637
|   |-- yolo26n.axmodel
|   |-- yolo26s.axmodel
|-- install_650
|   |-- ax_yolo26
|-- bus.jpg
|-- yolo26_out.jpg

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

(base) root@ax650:~# ./ax_yolo26 -m ./ax650/yolo26n.axmodel -i BUS.JPG
--------------------------------------
model file : ./ax650/yolo26n.axmodel
image file : BUS.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.
------------------------------------
Repeat 1 times, avg time 1.38 ms, max_time 1.38 ms, min_time 1.38 ms
--------------------------------------
detection num: 5
 5:  94%, [   6,  233,  801,  752], bus
 0:  94%, [  51,  396,  241,  904], person
 0:  91%, [ 227,  406,  345,  861], person
 0:  80%, [ 670,  389,  809,  876], person
 0:  50%, [   0,  556,   64,  872], person
--------------------------------------

Output image:

Downloads last month
111
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support