Scaled-YOLOv4: Scaling Cross Stage Partial Network

Code

• github.com/WongKinYiu/ScaledYOLOv4
  Officialpytorch★ 2,028
• github.com/AlexeyAB/darknet
  tf★ 22,190
• github.com/RangiLyu/nanodet
  pytorch★ 6,176
• github.com/Ededu1984/car_detection_forza_horizon
  tf★ 81
• github.com/Eric3911/ScaledYOLOv4
  pytorch★ 56
• github.com/wangermeng2021/ScaledYOLOv4-tensorflow2
  tf★ 47
• github.com/adam123wu/edge-yolo
  pytorch★ 14
• github.com/ikuokuo/start-scaled-yolov4
  pytorch★ 10
• github.com/achen353/Taiwanese-Traffic-Object-Detection
  tf★ 9
• github.com/leggedrobotics/darknet
  pytorch★ 9

Abstract

We show that the YOLOv4 object detection neural network based on the CSP approach, scales both up and down and is applicable to small and large networks while maintaining optimal speed and accuracy. We propose a network scaling approach that modifies not only the depth, width, resolution, but also structure of the network. YOLOv4-large model achieves state-of-the-art results: 55.5% AP (73.4% AP50) for the MS COCO dataset at a speed of ~16 FPS on Tesla V100, while with the test time augmentation, YOLOv4-large achieves 56.0% AP (73.3 AP50). To the best of our knowledge, this is currently the highest accuracy on the COCO dataset among any published work. The YOLOv4-tiny model achieves 22.0% AP (42.0% AP50) at a speed of 443 FPS on RTX 2080Ti, while by using TensorRT, batch size = 4 and FP16-precision the YOLOv4-tiny achieves 1774 FPS.

Tasks

Benchmark Results

Reproductions