This is the official implementation of our paper titled "ETDNet: Efficient Transformer-Based Detection Network for Surface Defect Detection". This paper has been accepted by IEEE Transactions on Instrumentation and Measurement (IEEE TIM).
For more details, please refer to our paper. This repo is based on PyTorch.
We release NEU-DET dataset used in this paper, including VOC style and COCO style. Note that five-flod cross-validation is used to divide the dataset.
Download the dataset(s) from corresponding links below.
- NEU-DET: Google Drive; BaiduNetDisk
The COCO pretrained weight of our ETDNet:
- ETDNet Weight: Google Drive; BaiduNetDisk
More configuration please see tools/mytrain.py and exps/neu_det/etdnet_neu_flod1.py, where you can set device num, dataset path, batch size, weight path, and so on.
python3 tools/mytrain.py -f exps/neu_det/etdnet_neu_flod1.py -d 4 -b 16 -c weigth/etdnet.pth --fp16
Our ETDNet achieves excellent performance on the NEU-DET dataset and is more lightweight and efficient.
The detection accuracy of our method is much higher than that of the yolo series.
Examples of detection results obtained by the proposed ETDNet on NEU-DET dataset. (a) Ground truth; (b) Prediction images by ETDNet.
Please cite our work if you find our work and codes helpful for your research.
@article{zhou2023etdnet,
title={ETDNet: Efficient Transformer-based Detection Network for Surface Defect Detection},
author={Zhou, Hantao and Yang, Rui and Hu, Runze and Shu, Chang and Tang, Xiaochu and Li, Xiu},
journal={IEEE Transactions on Instrumentation and Measurement},
year={2023},
publisher={IEEE}
}
This project is built upon numerous previous projects. We'd like to thank the contributors of YOLOX.