Verdict

For 30 FPS on a Jetson Orin Nano Super, D-FINE-S is the most accurate model: 53.4 mAP at 30.2 FPS, Apache-2.0. For 60 FPS, YOLO26-N gives 40.1 mAP (AGPL-3.0). The best Apache-2.0 option at that rate is DEIMv2-Femto, 34.5 mAP at 65.7 FPS. D-FINE-X is the most accurate model overall at 61.4 mAP, but it runs at 14.2 FPS. All numbers are TensorRT FP16, batch 1, mAP@50-95 in percent.

Fastest models on the Orin Nano Super

#ModelmAP@50-95FPSms/imageInput (px)Params (M)License
1DEIMv2-Atto27.589.111.233200.5Apache-2.0
2DEIMv2-Femto34.565.715.234161.0Apache-2.0
3YOLO26-N40.160.216.616402.4AGPL-3.0
4YOLOv5nu33.856.317.766402.6AGPL-3.0
5YOLO11-N38.752.619.006402.6AGPL-3.0
6YOLOX-Tiny35.551.619.394165.1Apache-2.0
7YOLOX-Nano28.851.019.624160.9Apache-2.0
8YOLOv8-N36.749.520.216403.2AGPL-3.0
9DEIMv2-Pico42.344.522.506401.5Apache-2.0
10YOLOv5su42.442.823.396409.1AGPL-3.0
11D-FINE-N45.842.123.776404.0Apache-2.0
12DEIM-N46.841.823.916404.0Apache-2.0
13YOLO26-S47.441.723.976409.5AGPL-3.0
14YOLO11-S45.741.324.206409.4AGPL-3.0
15DEIMv2-N46.641.224.256403.6Apache-2.0
Fastest detectors on Jetson Orin Nano Super, TensorRT FP16, batch 1. Check the input column: some top rows run on smaller images.

DEIMv2-Atto tops this table at 89.1 FPS, but it runs at 320 px. DEIMv2-Femto, YOLOX-Tiny and YOLOX-Nano run at 416 px. Most other models run at 640 px. If you came for YOLO11: YOLO11-N runs 52.6 FPS at 38.7 mAP and YOLO11-S 41.3 FPS at 45.7 mAP, both AGPL-3.0. D-FINE-N matches YOLO11-S at 45.8 mAP and 42.1 FPS under Apache-2.0.

Most accurate models

#ModelmAP@50-95FPSms/imageInput (px)Params (M)License
1D-FINE-X61.414.270.4564062.0Apache-2.0
2ECDet-X61.05.7174.8164049.0Apache-2.0
3DEIMv2-X60.75.4184.5464050.3Apache-2.0
4ECDet-L60.06.2161.5264031.0Apache-2.0
5D-FINE-L60.018.653.6464031.0Apache-2.0
6RT-DETRv4-X60.013.872.6964062.0Apache-2.0
7DEIM-X59.614.170.9364062.0Apache-2.0
8RF-DETR-L58.78.4119.33704128.0Apache-2.0
9DEIMv2-L58.56.2160.8164032.2Apache-2.0
10ECDet-M58.38.3119.7064018.0Apache-2.0
Most accurate detectors on Jetson Orin Nano Super, TensorRT FP16, batch 1.

The top of this table splits by speed. D-FINE-X reaches 61.4 mAP at 14.2 FPS. ECDet-X and DEIMv2-X land within a point of it but run at 5.7 and 5.4 FPS. D-FINE-L, at 60.0 mAP and 18.6 FPS, gives up just over a point for about a third more frame rate.

Every number on this page comes from the verified dataset: same 500-image COCO val2017 slice, conf 0.001, IoU 0.6, max 300 detections, pycocotools mAP, identical protocol across all hardware and runtimes. The full protocol is on the methodology page. To rerun this comparison with your own filters, open compare. Accuracy is measured on LibreYOLO retrained checkpoints; other weight sources can yield different values.