Home

Awesome

Interior Attention-Aware Network for Infrared Small Target Detection

Official implementation for IEEE Transactions on Geoscience and Remote Sensing (TGRS) paper: "Interior Attention-Aware Network for Infrared Small Target Detection". [Paper]

News

[2023/12/11] I plan to reorganize the code and provide a more friendly model version.

Requirement

Packages:

File Structure

iaanet
├─ box_generate.py
├─ cGAN_data
│  └─ training_box_gt.csv
├─ detect.py
├─ models
│  ├─ attention.py
│  ├─ backbone.py
│  ├─ embedding.py
│  └─ transformer.py
├─ pretrained
│  ├─ iaanet.pt
│  └─ rpn.pt
├─ test.py
├─ train.py
└─ utils
   ├─ datasets.py
   ├─ general.py
   └─ loss.py

Dataset Preparation

cGAN_data
├─ training
│  ├─ 000000_1.png
│  ├─ 000000_2.png
│  ├─ ...
│  ├─ 009999_1.png
│  └─ 009999_2.png
├─ test_org
│  ├─ 00000.png
│  ├─ ...
│  └─ 00099.png 
└─  test_gt
   ├─ 00000.png
   ├─ ...
   └─ 00099.png 
cGAN_data
├─ ...
├─ val_org
└─ val_gt
cGAN_data
├─ ...
└─ training_box_gt.csv
python box_generate.py --path ./cGAN_data/training/ --save_path ./cGAN_data/training_box_gt.csv --bord 4

Training

Experiments are conducted using PyTorch with a single GeForce RTX 3090 GPU of 24 GB Memory.

Train from scratch

python train.py --batch_size 8 --epochs 10 --save_path ./outputs/demo/

Start from pretraind RPN

python train.py --rpn_pretrained ./pretrained/rpn.pt --save_path ./outputs/demo/

Run python train.py --help for more configurations

Testing

Use pretrained model for testing

python test.py --weights ./pretrained/iaanet.pt

Fast version (SG convs the proposed regions only. ):

python test.py --weights ./pretrained/iaanet.pt --fast

Run python test.py --help for more configurations

Results

We follow MDvsFA-cGAN to calculate F-measure [Code]

DatasetF-measurePrecisionRecall
MDvsFA0.6390.6060.818

Inference

Infer a single image

python detect.py --image_path img.png --save_path ./inference/ --weights ./pretrained/iaanet.pt

Infer images in a folder

python detect.py --image_path ./folder/ --save_path ./inference/ --weights ./pretrained/iaanet.pt --folder

Fast version

python detect.py --fast

Run python detect.py --help for more configurations

Citation

If you find our work useful in your research, please cite our paper

@ARTICLE{9745054,
  author={Wang, Kewei and Du, Shuaiyuan and Liu, Chengxin and Cao, Zhiguo},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Interior Attention-Aware Network for Infrared Small Target Detection}, 
  year={2022},
  doi={10.1109/TGRS.2022.3163410}
  }

License

MIT License