Submission details of YOLOv3

Name YOLOv3
Paper Link https://arxiv.org/abs/1804.02767
Code Link https://github.com/eriklindernoren/PyTorch-YOLOv3
AP 0.417308
Input N/A
Runtime 0.051 s
Environment 1 GPU (Titan X)
Abstract We present some updates to YOLO! We made a bunch of little design changes to make it better. We also trained this new network that’s pretty swell. It’s a little bigger than last time but more accurate. It’s still fast though, don’t worry. At 320 × 320 YOLOv3 runs in 22 ms at 28.2 mAP, as accurate as SSD but three times faster. When we look at the old .5 IOU mAP detection metric YOLOv3 is quite good. It achieves 57.9 AP50 in 51 ms on a Titan X, compared to 57.5 AP50 in 198 ms by RetinaNet, similar performance but 3.8× faster.

Detailed results

Overall precision/recall curve

Per-sequence results

sequence name AP
cubberly-auditorium-2019-04-22_1 0.52121
discovery-walk-2019-02-28_0 0.542249
discovery-walk-2019-02-28_1 0.526636
food-trucks-2019-02-12_0 0.589604
gates-ai-lab-2019-04-17_0 0.620315
gates-basement-elevators-2019-01-17_0 0.635444
gates-foyer-2019-01-17_0 0.730365
gates-to-clark-2019-02-28_0 0.681083
hewlett-class-2019-01-23_0 0.724606
hewlett-class-2019-01-23_1 0.819074
huang-2-2019-01-25_1 0.338966
huang-intersection-2019-01-22_0 0.516353
indoor-coupa-cafe-2019-02-06_0 0.37629
lomita-serra-intersection-2019-01-30_0 0.500425
meyer-green-2019-03-16_1 0.320416
nvidia-aud-2019-01-25_0 0.327894
nvidia-aud-2019-04-18_1 0.646028
nvidia-aud-2019-04-18_2 0.303314
outdoor-coupa-cafe-2019-02-06_0 0.323565
quarry-road-2019-02-28_0 0.496468
serra-street-2019-01-30_0 0.326028
stlc-111-2019-04-19_1 0.707106
stlc-111-2019-04-19_2 0.645326
tressider-2019-03-16_2 0.418646
tressider-2019-04-26_0 0.263253
tressider-2019-04-26_1 0.3811
tressider-2019-04-26_3 0.348149
total 0.417308