Scaled-YOLOv4: Scaling Cross Stage Partial Network

TL;DR

This paper introduces a scalable version of YOLOv4 that can be adapted for different sizes and applications, achieving state-of-the-art accuracy and speed on the COCO dataset through a novel network scaling approach.

Contribution

It proposes a comprehensive network scaling method that adjusts depth, width, resolution, and structure, enabling YOLOv4 to perform optimally across various sizes.

Findings

YOLOv4-large achieves 55.5% AP on COCO at 16 FPS.

YOLOv4-tiny reaches 22.0% AP at 443 FPS.

Scaling improves accuracy and speed across different network sizes.

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.