RTMDet – PyTorch Implementation

This repository is a PyTorch port of RTMDet, originally implemented in MMDetection.

The goal is to reimplement the network in pure PyTorch while making it possible to load pretrained weights from the original models.

Installation

pip install rtmdet

Usage

from rtmdet import RTMDet

model = RTMDet.from_preset("small")  # tiny / small / medium / large
bboxes, scores, classes = model("image.jpg")

References

• RTMDet: An Empirical Study of Real-Time Object Detectors
  Xiangyu Zhang, Xinyu Zhou, Zhiqi Li, et al.
  📄 Paper

Acknowledgments

Based on MMDetection and the official RTMDet implementation.