RTMDet: An Empirical Study of Designing Real-Time Object Detectors

TL;DR

RTMDet is a new real-time object detector that outperforms existing models like YOLO in accuracy and speed, with versatile architecture suitable for various object recognition tasks, including segmentation and rotated detection.

Contribution

The paper introduces RTMDet, a highly efficient and extensible real-time object detector with a novel architecture using large-kernel depth-wise convolutions and improved training techniques.

Findings

Achieves 52.8% AP on COCO at 300+ FPS.

Outperforms mainstream industrial detectors in accuracy and speed.

Sets new state-of-the-art in real-time instance segmentation and rotated detection.

Abstract

In this paper, we aim to design an efficient real-time object detector that exceeds the YOLO series and is easily extensible for many object recognition tasks such as instance segmentation and rotated object detection. To obtain a more efficient model architecture, we explore an architecture that has compatible capacities in the backbone and neck, constructed by a basic building block that consists of large-kernel depth-wise convolutions. We further introduce soft labels when calculating matching costs in the dynamic label assignment to improve accuracy. Together with better training techniques, the resulting object detector, named RTMDet, achieves 52.8% AP on COCO with 300+ FPS on an NVIDIA 3090 GPU, outperforming the current mainstream industrial detectors. RTMDet achieves the best parameter-accuracy trade-off with tiny/small/medium/large/extra-large model sizes for various application scenarios, and obtains new state-of-the-art performance on real-time instance segmentation and rotated object detection. We hope the experimental results can provide new insights into designing versatile real-time object detectors for many object recognition tasks. Code and models are released at https://github.com/open-mmlab/mmdetection/tree/3.x/configs/rtmdet.