Sewer-ML: A Multi-Label Sewer Defect Classification Dataset and\n Benchmark

2021/03/19 by Joakim Bruslund Haurum, Haurum, Joakim Bruslund, Thomas B. Moeslund +1 · 4 citations
Computer Science · Engineering · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Infrastructure Maintenance and Monitoring #Music and Audio Processing #Water Systems and Optimization

paper · pdf · doi:10.48550/arxiv.2103.10895

openalex publication_date 2021/03/19 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/30

Abstract

Perhaps surprisingly sewerage infrastructure is one of the most costly\ninfrastructures in modern society. Sewer pipes are manually inspected to\ndetermine whether the pipes are defective. However, this process is limited by\nthe number of qualified inspectors and the time it takes to inspect a pipe.\nAutomatization of this process is therefore of high interest. So far, the\nsuccess of computer vision approaches for sewer defect classification has been\nlimited when compared to the success in other fields mainly due to the lack of\npublic datasets. To this end, in this work we present a large novel and\npublicly available multi-label classification dataset for image-based sewer\ndefect classification called Sewer-ML.\n The Sewer-ML dataset consists of 1.3 million images annotated by professional\nsewer inspectors from three different utility companies across nine years.\nTogether with the dataset, we also present a benchmark algorithm and a novel\nmetric for assessing performance. The benchmark algorithm is a result of\nevaluating 12 state-of-the-art algorithms, six from the sewer defect\nclassification domain and six from the multi-label classification domain, and\ncombining the best performing algorithms. The novel metric is a\nclass-importance weighted F2 score, \F2_\CIW, reflecting the\neconomic impact of each class, used together with the normal pipe F1 score,\n\F1_\Normal. The benchmark algorithm achieves an\n\F2_\CIW score of 55.11% and \F1_\Normal score\nof 90.94%, leaving ample room for improvement on the Sewer-ML dataset. The\ncode, models, and dataset are available at the project page\nhttps://vap.aau.dk/sewer-ml/\n

Citations

• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• The Pascal Visual Object Classes (VOC) Challenge
• A Review on Multi-Label Learning Algorithms
• ImageNet Large Scale Visual Recognition Challenge
• ImageNet classification with deep convolutional neural networks
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal\n Networks
• Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
• Detecting Functionality-Specific Vulnerabilities via Retrieving Individual Functionality-Equivalent APIs in Open-Source Repositories
• Asymmetric Loss For Multi-Label Classification
• Deep Learning for Anomaly Detection: A Review

Cited by

• Sewer pipeline condition assessment and defect detection using computer vision
• Explainable Deep Anomaly Detection with Sequential Hypothesis Testing for Robotic Sewer Inspection
• Multi-scale hybrid vision transformer and Sinkhorn tokenizer for sewer defect classification
• Real-time defect detection in underground sewage pipelines using an improved YOLOv5 model

Related

• Multi-label Sewer Pipe Defect Recognition with Mask Attention Feature Enhancement and Label Correlation Learning
• Towards Trustworthy Multi-label Sewer Defect Classification via Evidential Deep Learning
• Sewer Image Super-Resolution with Depth Priors and Its Lightweight Network
• Self-Supervised Learning for Identifying Defects in Sewer Footage
• When the Small-Loss Trick is Not Enough: Multi-Label Image Classification with Noisy Labels Applied to CCTV Sewer Inspections