YOLOv3: An Incremental Improvement

TL;DR

YOLOv3 introduces incremental improvements to the YOLO object detection model, achieving higher accuracy while maintaining real-time speed, making it competitive with other state-of-the-art detectors.

Contribution

The paper presents design modifications and a new trained network that improves YOLO's accuracy without sacrificing its real-time performance.

Findings

YOLOv3 runs in 22 ms at 28.2 mAP on 320x320 images.

It achieves 57.9 mAP@50 in 51 ms on a Titan X.

Compared to RetinaNet, it is similarly accurate but 3.8 times faster.

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 320x320 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 mAP@50 in 51 ms on a Titan X, compared to 57.5 mAP@50 in 198 ms by RetinaNet, similar performance but 3.8x faster. As always, all the code is online at https://pjreddie.com/yolo/