SSTNet

The Pytorch implementation of the TGRS 2024 paper SSTNet: Sliced spatio-temporal network with cross-slice ConvLSTM for moving infrared dim-small target detection

Abstract

Infrared dim-small target detection, as an important branch of object detection, has been attracting research attention in recent decades. Its challenges mainly lie in the small target sizes and dim contrast to background images. Recent research schemes on it mainly focus on improving the feature representation of spatio-temporal domains only in single-slice temporal scope. More cross-slice motion, i.e., past and future, is seldom considered to enhance target features. To utilize cross-slice motion context, this paper proposes a sliced spatio-temporal network (SSTNet) with cross-slice enhancement for moving infrared dim-small target detection. In our scheme, a new cross-slice ConvLSTM node is designed to capture spatio-temporal motion features from both inner-slice and inter-slices. Moreover, to improve infrared small target motion feature learning, we extend conventional loss function by adopting a new motion-coordination loss term. On these, we propose a motion-coupling neck to assist feature extractor in facilitating the capturing and utilization of motion features from multi-frames. To our best knowledge, our work is the first one to explore the cross-slice spatio-temporal motion modeling for infrared dim-small targets. Experiments verify that our SSTNet could refresh most state-of-the-art metrics on two public benchmarks (DAUB and IRDST).

Datasets

• Datasets are available at DAUB(code: saew) and IRDST(code: rrnr).

• Or you can download them directly from the website: DAUB and IRDST. In addition, you need to reorganize these datasets in a format similar to the coco_train.txt and coco_val.txt file we provided.

• The COCO format json file needs to be converted into a txt format file. We provide txt format files for DAUB and IRDST.

• Or you can generate a new txt file based on the path of your datasets. We also provide json files for DAUB and IRDST.

python utils_coco/coco_to_txt.py

Prerequisite

• python==3.10.11
• pytorch==1.12.0
• torchvision==0.13.0
• numpy==1.24.3
• opencv-python==4.7.0.72
• pillow==9.5.0
• scipy==1.10.1
• Tested on Ubuntu 20.04, with CUDA 11.3, and 1x NVIDIA 3090.

Usage

Train

CUDA_VISIBLE_DEVICES=0 python train_{dataset}.py

Test

• Usually model_best.pth is not necessarily the best model. The best model may have a lower val_loss or a higher AP50 during verification.

python vid_map_coco.py

Visulization

python vid_predict.py

Results

• Pre-trained models

Method	Dataset	mAP50 (%)	Precision (%)	Recall (%)	F1 (%)	Download
SSTNet	DAUB	95.59	98.08	98.10	98.09	models (code: 68sq)

• PR curves on DAUB and IRDST datasets.
• We provide the results on DAUB and IRDST, and you can plot them using Python.

Contact

If any questions, kindly contact with Shengjia Chen via e-mail: [email protected].

References

1. X. Shi, Z. Chen, H. Wang, D.-Y. Yeung, W.-K. Wong, and W.-c. Woo, “Convolutional lstm network: A machine learning approach for precipitation nowcasting,” Advances in Neural Information Processing Systems, vol. 28, 2015.
2. Z. Ge, S. Liu, F. Wang, Z. Li, and J. Sun, “Yolox: Exceeding yolo series in 2021,” arXiv preprint arXiv:2107.08430, 2021.

Citation

If you find this repo useful, please cite our paper.

@ARTICLE{sstnet2024,
  author={Chen, Shengjia and Ji, Luping and Zhu, Jiewen and Ye, Mao and Yao, Xiaoyong},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={SSTNet: Sliced Spatio-Temporal Network With Cross-Slice ConvLSTM for Moving Infrared Dim-Small Target Detection}, 
  year={2024},
  volume={62},
  pages={1-12},
  doi={10.1109/TGRS.2024.3350024}}