LangGas: Introducing Language in Selective Zero-Shot Background Subtraction for Semi-Transparent Gas Leak Detection with a New Dataset

2025-03-04

AI Summary

Detecting semi-transparent gas leaks remains challenging due to their low visual contrast and the absence of high-quality, publicly available datasets. To address this, we propose a language-guided zero-shot background subtraction framework that enables end-to-end detection without requiring real leakage samples. Our method integrates language prompting (via GLIP), adaptive background modeling, multi-stage semantic filtering, and prompt-driven threshold optimization. Furthermore, we introduce the first synthetic benchmark dataset for gas leak detection—generated using Blender+PhysX—featuring diverse backgrounds, realistic occluders, and pixel-level segmentation annotations. Experiments demonstrate that our approach achieves 69% mIoU on the new benchmark, significantly outperforming conventional background subtraction and zero-shot detection baselines, while exhibiting strong cross-scene generalization. Both code and dataset will be made publicly available.

Technology Category

• Natural Language Processing: Language Grounding & Multi-modal NLP
• Computer Vision: Language and Vision
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Security and Privacy: Data transparency and provenance
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Gas leakage poses a significant hazard that requires prevention. Traditionally, human inspection has been used for detection, a slow and labour-intensive process. Recent research has applied machine learning techniques to this problem, yet there remains a shortage of high-quality, publicly available datasets. This paper introduces a synthetic dataset featuring diverse backgrounds, interfering foreground objects, diverse leak locations, and precise segmentation ground truth. We propose a zero-shot method that combines background subtraction, zero-shot object detection, filtering, and segmentation to leverage this dataset. Experimental results indicate that our approach significantly outperforms baseline methods based solely on background subtraction and zero-shot object detection with segmentation, reaching an IoU of 69% overall. We also present an analysis of various prompt configurations and threshold settings to provide deeper insights into the performance of our method. The code and dataset will be released after publication.

Problem

• Detects semi-transparent gas leaks using zero-shot learning.
• Addresses lack of high-quality datasets for gas leak detection.
• Improves detection accuracy with advanced segmentation techniques.

Innovation

• Synthetic dataset with precise segmentation ground truth
• Zero-shot method combining background subtraction and detection
• Analysis of prompt configurations and threshold settings