countingstars619/scratchdet

The code and models for paper: "ScratchDet: Exploring to Train Single-Shot Object Detectors from Scratch"

ScratchDet: Training Single-Shot Object Detectors from Scratch

By Rui Zhu*, Shifeng Zhang*, Xiaobo Wang, Longyin Wen, Hailin Shi†, Liefeng Bo, Tao Mei. (*Equal Contribution, †Corresponding author)

The code is originally based on the SSD-caffe and RefineDet-caffe framework. We also implement on mmdetection. If you want to use one kind of codes, please follow their instructions to finish the initial install.

Please cite our paper in your publications if it helps your research:

@inproceedings{zhu2019scratchdet,
  title={ScratchDet: Training Single-Shot Object Detectors From Scratch},
  author={Zhu, Rui and Zhang, Shifeng and Wang, Xiaobo and Wen, Longyin and Shi, Hailin and Bo, Liefeng and Mei, Tao},
  booktitle={CVPR},
  year={2019}
}

Introduction

ScratchDet focus on training object detectors from scratch in order to tackle the problems caused by fine-tuning from pretrained networks. In this paper, we study the effects of BatchNorm in the backbone and detection head subnetworks, and successfully train detectors from scratch. By taking the pretaining-free advantage, we are able to explore various architectures for detector designing. Please see our paper for more details.

Figure : Gradient Analysis.

Codes

You can see details and codes in the folder caffe/ and mmdetection/ .

Contact

Rui Zhu (zhur5 at mail2.sysu.edu.cn)

Any comments or suggestions are welcome!