Invisible Gas Detection: An RGB-Thermal Cross Attention Network and A New Benchmark

Accepted to CVIU (JCR Q1, CCF rank B) 🎉🎉🥳🥳

Paper (Latest version) | Gas-DB

Authors

• Jue Wang*^1,3, Yuxiang Lin*², Qi Zhao³, Dong Luo³, Shuaibao Chen³, Wei Chen³, Xiaojiang Peng²
• ¹Southern University of Science and Technology, ²Shenzhen Technology University, ³Shenzhen Institute of Technology, CAS [Jue and Yuxiang contribute equally to this work.]

If you are interested in our work, please star ⭐ our project.

Training dataset preparation

• Prepare our Gas-DB dataset: please download in Gas-DB.

Code

Setup

conda create -n RT-CAN python==3.8.16
conda activate RT-CAN
pip install -r requirements.txt

Train

python train.py

Test

python test.py

Infetence

python inference.py

RGB-Thermal Cross Attention Network

Illustration the architecture of RGB-Thermal Two Stream Cross Attention Network. (a) Two stream RGB-ThermaR Cl Encoder, (b) Cascaded Decoder.

An overview of our Gas-DB  

This figure shows an overview of our Gas-DB, containing 8 kinds of scenery, containing sunny, rainy, double leakage, nearly leakage, further leakage, overlook, simple background, and complex background. The last one is the original gas image without manually annotating.

The comparision of the GasVid and our Gas-DB.

The visualization of the prediction comparisons from different methods

The visualization of the prediction comparisons from different methods, according to the rows from top to bottom in order: RGB; Thermal; Ground Truth; PSPNet; Segformer; YOLOv5; MFNet; EAEFNet; Ours.

Contact   

For any question, feel free to email j.wang2@siat.ac.cn and yuxiang.lin@gatech.edu.

Citation

@article{RT-CAN,
title = {Invisible gas detection: An RGB-thermal cross attention network and a new benchmark},
journal = {Computer Vision and Image Understanding},
volume = {248},
pages = {104099},
year = {2024},
issn = {1077-3142},
doi = {https://doi.org/10.1016/j.cviu.2024.104099},
url = {https://www.sciencedirect.com/science/article/pii/S1077314224001802},
author = {Jue Wang and Yuxiang Lin and Qi Zhao and Dong Luo and Shuaibao Chen and Wei Chen and Xiaojiang Peng},
keywords = {Gas detection, Computer vision, RGB-Thermal, Gas-DB},
abstract = {The widespread use of various chemical gases in industrial processes necessitates effective measures to prevent their leakage during transportation and storage, given their high toxicity. Thermal infrared-based computer vision detection techniques provide a straightforward approach to identify gas leakage areas. However, the development of high-quality algorithms has been challenging due to the low texture in thermal images and the lack of open-source datasets. In this paper, we present the RGB-Thermal Cross Attention Network (RT-CAN), which employs an RGB-assisted two-stream network architecture to integrate texture information from RGB images and gas area information from thermal images. Additionally, to facilitate the research of invisible gas detection, we introduce Gas-DB, an extensive open-source gas detection database including about 1.3K well-annotated RGB-thermal images with eight variant collection scenes. Experimental results demonstrate that our method successfully leverages the advantages of both modalities, achieving state-of-the-art (SOTA) performance among RGB-thermal methods, surpassing single-stream SOTA models in terms of accuracy, Intersection of Union (IoU), and F2 metrics by 4.86%, 5.65%, and 4.88%, respectively. The code and data can be found at https://github.com/logic112358/RT-CAN.}
}