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: 5.0

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

• AX650N/AX8850
  • M4N-Dock(爱芯派Pro)
  • M.2 Accelerator card
• AX630C
  • 爱芯派2
  • Module-LLM
  • LLM630 Compute Kit
• AX615
• AX637

Performance Statistics

AX650N

Model	Latency(ms) npu1	Latency(ms) npu3
yolo26n-seg	5.096	1.847
yolo26s-seg	13.724	4.735
yolo26m-seg	34.582	11.912
yolo26l-seg	65.275	22.312
yolo26x-seg	110.208	36.147

AX630C

Model	Latency(ms) npu1	Latency(ms) npu2
yolo26n-seg	18.561	12.036
yolo26s-seg	38.493	30.259
yolo26m-seg	94.779	69.217
yolo26l-seg	179.964	121.261
yolo26x-seg	296.999	190.418

AX615

Model	Latency(ms) npu1	Latency(ms) npu2
yolo26n-seg	22.024	13.867
yolo26s-seg	62.068	35.333
yolo26m-seg	160.450	86.186

AX637

Model	Latency(ms) npu1
yolo26n-seg	6.175
yolo26s-seg	15.654
yolo26m-seg	35.612
yolo26l-seg	67.062
yolo26x-seg	104.862

How to use

Download all files from this repository to the device

Inference

Input image:

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

(base) root@ax650:~/ax650seg# python3 ax_infer.py --model-path yolov8m-seg_640x640_npu3.axmodel --test-img bus.jpg
[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-Seg] [15:38:19.398] [DEBUG] Load model time = 592.02 ms
[YOLOv8-Seg] [15:38:19.433] [DEBUG] Pre-process time = 8.17 ms
[YOLOv8-Seg] [15:38:19.472] [DEBUG] Forward time = 38.19 ms
[YOLOv8-Seg] [15:38:19.482] [DEBUG] Post-process time = 9.47 ms
[YOLOv8-Seg] [15:38:19.483] [DEBUG] Proto shape: (32, 160, 160)
[YOLOv8-Seg] [15:38:19.534] [INFO] Draw Results (6 objects):
[YOLOv8-Seg] [15:38:19.535] [INFO] (13, 230, 803, 736) -> bus: 0.93
[YOLOv8-Seg] [15:38:19.580] [INFO] (49, 398, 244, 904) -> person: 0.91
[YOLOv8-Seg] [15:38:19.599] [INFO] (222, 397, 345, 860) -> person: 0.89
[YOLOv8-Seg] [15:38:19.616] [INFO] (667, 395, 810, 880) -> person: 0.88
[YOLOv8-Seg] [15:38:19.632] [INFO] (0, 551, 78, 866) -> person: 0.59
[YOLOv8-Seg] [15:38:19.647] [INFO] (137, 472, 148, 504) -> tie: 0.37
[YOLOv8-Seg] [15:38:19.685] [INFO] Saved to result_yolov8_seg.jpg

Output image: