TAMNet: Triple Adaptive Multiplexing Network for Wide-Area Infrared Small Target Detection

点击这里查看文章

Requirements

Python 3
pytorch 1.2.0 or higher
numpy, PIL, tqdm, shutil

Algorithm Introduction

We propose a triple adaptive multiplexing network (TAMNet) for wide-area infrared small target detection (WIRSTD) in this paper. Experiments on the dataset provided by the PRCV2024 Wide Area Infrared Small Target Detection Challenge (hereafter PRCV2024) demonstrate the effectiveness of our method. The contributions can be summarized as follows:

1. We propose TAMNet featuring dedicated adaptive multiplexing strategies in feature extraction, contextual understanding, and prediction generation to achieve exceptional cross-domain generalization for WIRSTD.
   
2. We propose a novel attention-decoupled parallel attention residual block (PARB) module that adaptively integrates multi-branch features to maximally preserve and reinforce subtle features while effectively mitigating cross-domain feature distortion induced by attention mechanisms.
   
3. We propose local saliency module (LSM) as a replacement for the skip connections (SK) to handle complex background variations by explicitly modeling multi-scale context to implicitly derive stable targetbackground relationships.
   

Datasets

We used the PRCV2024 for both training and test. Please place the dataset to the folder ./datasets/. Our project has the following structure:

├──./datasets/
│    ├── PRCV2024
│    │    ├── images
│    │    │    ├── XXX.png
│    │    │    ├── ...
│    │    ├── masks
│    │    │    ├── xxx.png
│    │    │    ├── ...
│    │    ├── img_idx
│    │    │    ├── train_PRCV2024.txt
│    │    │    ├── test_PRCV2024.txt
│    │    │    ├── validation_PRCV2024.txt

Commands

Commands for train

Run trainVMDNAL.py to perform network training in single GPU and multiple GPUs.

Checkpoints and Logs will be saved to ./log/, and the ./log/ has the following structure:

├──./log/
│    ├── PRCV2024 
│    │    ├── VMDNAL_eopch400.pth.tar

Commands for test

Run testVMDNAL.py to perform network inference and evaluation. The PA/mIoU and PD/FA values will be saved to ./test_[current time].txt Network predictions will be saved to ./results/ that has the following structure:

├──./results/
│    ├── PRCV2024
│    │   ├── VMDNAL
│    │   │    ├── XXX.png

Recources

The pre-trained models can be downlaod via Baidu Drive (key:8mj5) and One Drive.

Acknowledgement

The overall repository style is highly borrowed from BasicIRSTD.

Citation

If you find the code useful, please consider citing our paper using the following BibTeX entry.

@article{WANG2026133323,
title = {TAMNet: Triple adaptive multiplexing network for wide-area infrared small target detection},
journal = {Neurocomputing},
volume = {681},
pages = {133323},
year = {2026},
issn = {0925-2312},
author = {Ru Wang and Fanglong Wu and Yadong Chen and Peng Cheng},
}