Rethinking Generalizable Infrared Small Target Detection: A Real-scene Benchmark and Cross-view Representation Learning
- Introduction
- Dependencies and Installation
- Dataset
- Codes Demos
- Usage
- Quantative Results
- Citations
- Acknowledgement
- Following SCTransNet
- Python == 3.8
- pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
Download the dataset RealScene-ISTD：Download dir[Baidu Drive].
- The following structure:
├──./datasets/
│ ├── RealScene-ISTD
│ │ ├── images
│ │ │ ├── 000001.png
│ │ │ ├── 000002.png
│ │ │ ├── ...
│ │ ├── masks
│ │ │ ├── 000001.png
│ │ │ ├── 000002.png
│ │ │ ├── ...
│ │ ├── img_idx
│ │ │ ├── train_RealScene-ISTD.txt
│ │ │ ├── test_RealScene-ISTD.txt
Gamma_Correction
# Run gamma_correction.py directly
python Dataset_Alignment/gamma_correction.pyTopk_Fusion
# Run Topk_Fusion directly
python Dataset_Alignment/Topk_Fusion.pyYou can use our pre-prepared small target patches [Baidu Drive].
python train.pypython test.py| Method | mIoU (x10(-2)) | Pd (x10(-2)) | Fa (x10(-6)) | |
|---|---|---|---|---|
| ACM-Net | 64.25 | 91.97 | 250.79 | |
| ALC-Net | 68.88 | 90.91 | 146.85 | |
| DNA-Net | 73.79 | 91.33 | 60.06 | |
| RDIAN | 59.46 | 90.70 | 357.92 | |
| ISTDU-Net | 74.40 | 93.87 | 75.35 | |
| UIU-Net | 71.14 | 90.91 | 107.71 | |
| SCTransNet | 75.01 | 94.29 | 53.59 | |
| Ours | 79.32 | 96.83 | 5.40 | [Weights] |
| Method | mIoU (x10(-2)) | Pd (x10(-2)) | Fa (x10(-6)) | |
|---|---|---|---|---|
| ACM-Net | 57.68 | 92.23 | 115.28 | |
| ALC-Net | 62.69 | 92.28 | 58.99 | |
| DNA-Net | 64.14 | 95.64 | 40.90 | |
| RDIAN | 60.12 | 93.96 | 98.88 | |
| ISTDU-Net | 63.20 | 96.98 | 14.36 | |
| UIU-Net | 62.76 | 90.6 | 33.87 | |
| SCTransNet | 66.93 | 93.96 | 15.45 | |
| Ours | 72.44 | 95.30 | 14.08 | [Weights] |
This project is build based on SCTransNet. Thanks to Shuai Yuan.
The comparative experiments of our method with other methods are based on BasicIRSTD. Thanks to Xinyi Ying.