DMIST-Benchmark

Dense Moving Infrared Small Target Detection

The DMIST benchmark datasets and baseline model implementation of the TGRS 2024 paper Towards Dense Moving Infrared Small Target Detection: New Datasets and Baseline

Benchmark Datasets (bounding box-based)

• We synthesize two dense moving infrared small target datasets DMIST-60 and DMIST-100 on DAUB.

• Datasets are available at DMIST Baidu/Google and IRDST Baidu(code: t2ti). Or you can download IRDST directly from the website. In addition, we also introduce a new drone swarm dataset, DSISTDBaidu(code: r5cg), that integrates real targets into simulated infrared backgrounds.

• You need to reorganize these datasets in a format similar to the DMIST_train.txt and DMIST_val.txt files we provided (txt files are used in training). We provide the txt files for DMIST and IRDST. For example:

train_annotation_path = '/home/LASNet/DMIST_train.txt'
val_annotation_path = '/home/LASNet/DMIST_60_val.txt'

• Or you can generate a new txt file based on the path of your datasets. Text files (e.g., DMIST_60_val.txt) can be generated from json files (e.g., 60_coco_val.json). We also provide all json files for DMIST Baidu/Google and IRDST Baidu.

python utils_coco/coco_to_txt.py

• The folder structure should look like this:

DMIST
├─coco_train.json
├─60_coco_val.json
├─100_coco_val.json
├─images
│   ├─train
│   │   ├─data5
│   │   │   ├─0.bmp
│   │   │   ├─0.txt
│   │   │   ├─ ...
│   │   │   ├─2999.bmp
│   │   │   ├─2999.txt
│   │   │   ├─ ...
│   │   ├─ ...
│   ├─test60
│   │   ├─data6
│   │   │   ├─0.bmp
│   │   │   ├─0.txt
│   │   │   ├─ ...
│   │   │   ├─398.bmp
│   │   │   ├─398.txt
│   │   │   ├─ ...
│   │   ├─ ...
│   ├─test100
│   │   ├─ ...

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 of baseline LASNet

Train

CUDA_VISIBLE_DEVICES=0 python train_DMIST.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.

"model_path": '/home/LASNet/logs/model.pth'

• You need to change the path of the json file of test sets. For example:

#Use DMIST-100 dataset for test.
cocoGt_path         = '/home/public/DMIST/100_coco_val.json'
dataset_img_path    = '/home/public/DMIST/'

python test_DMIST.py

Visulization

• We support video and single-frame image prediction.

# mode = "video" #Predict a sequence
mode = "predict"  #Predict a single-frame image 

python predict.py

Results

• We optimize old codes and retrain LASNet, achieving slightly better performance results than those reported in our paper.

Method	Dataset	mAP50 (%)	Precision (%)	Recall (%)	F1 (%)	Download
LASNet	DMIST-60	76.47	95.84	80.07	87.25	Baidu (code: y7ki) Google
LASNet	DMIST-100	65.70	96.52	68.68	80.25

• PR curve on DMIST and IRDST datasets in the paper.
• We provide the results on DMIST-60, DMIST-100 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. S. Chen, L. Ji, J. Zhu, M. Ye and X. Yao, "SSTNet: Sliced Spatio-Temporal Network With Cross-Slice ConvLSTM for Moving Infrared Dim-Small Target Detection," in IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1-12, 2024, Art no. 5000912, doi: 10.1109/TGRS.2024.3350024.
2. B. Hui et al., “A dataset for infrared image dim-small aircraft target detection and tracking under ground/air background,” Sci. Data Bank, CSTR 31253.11.sciencedb.902, Oct. 2019.

Citation

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

@ARTICLE{chen2024dmist,
  author={Chen, Shengjia and Ji, Luping and Zhu, Sicheng and Ye, Mao and Ren, Haohao and Sang, Yongsheng},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Toward Dense Moving Infrared Small Target Detection: New Datasets and Baseline}, 
  year={2024},
  volume={62},
  pages={1-13},
  doi={10.1109/TGRS.2024.3443280}}