YOLO26 320x320 (nano and small)

Model Overview

YOLO26 object detection models optimized for 320x320 resolution. YOLO26 is Ultralytics' end-to-end NMS-free architecture, with the decode head compiled onto the NPU. This model page covers both the nano (yolo26n) and small (yolo26s) variants.

Both models are developed and optimized for the Synaptics Astra® SL2610-Series processors Torq NPU.

Model Features

• Model Type: Object Detection (NMS-free, one-to-one head)
• Input: 1x320x320x3 int8 NHWC (scale 1/255, zero-point -128)
• Output: 1x84x2100 int8. Rows 0-3 hold xywh normalized to the input size; rows 4-83 hold class probabilities, already decoded on the NPU. Int8 output quantization saturates confidences near 0.5, so choose the threshold accordingly.

Deployment

The compiled model files are available for download on Huggingface at Synaptics/yolov26n_od.

Usage tutorial available at Synaptics AI Developer Zone / SL2600 / Object Detection.

Performance on NPU

Model	Processor	Inference Time (ms)	FPS	COCO val2017 mAP50 (int8)
yolo26n	SL2610	12.9	78	0.403
yolo26s	SL2610	33.5	30	0.503

Post-processing

Dequantize the output, threshold the max class probability (about 0.25), scale xywh by the input size, and undo the letterbox. The NMS-free head needs no score-based NMS, but int8 confidence saturation produces tied duplicate boxes, so apply an IoU dedupe pass.

License

Both the source models and the compiled models for on-device deployment are licensed under AGPL-3.0.

Learn More

• Synaptics AI Developer Zone: Get started with documentation, tutorials and resources for your Edge AI journey.
• Astra Support Portal: Connect with our engineering team and community.