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

Perhaps surprisingly sewerage infrastructure is one of the most costly in modern society. Sewer pipes are manually inspected to whether the pipes are defective. However, this process is limited by number of qualified inspectors and the time it takes to inspect a pipe. of this process is therefore of high interest. So far, the of computer vision approaches for sewer defect classification has been when compared to the success in other fields mainly due to the lack of datasets. To this end, in this work we present a large novel and available multi-label classification dataset for image-based sewer classification called Sewer-ML. The Sewer-ML dataset consists of 1.3 million images annotated by professional inspectors from three different utility companies across nine years. with the dataset, we also present a benchmark algorithm and a novel for assessing performance. The benchmark algorithm is a result of 12 state-of-the-art algorithms, six from the sewer defect domain and six from the multi-label classification domain, and the best performing algorithms. The novel metric is a-importance weighted F2 score, 2_, reflecting the impact of each class, used together with the normal pipe F1 score,1_. The benchmark algorithm achieves an2_ score of 55.11% and 1_ score 90.94%, leaving ample room for improvement on the Sewer-ML dataset. The, models, and dataset are available at the project page://vap.aau.dk/sewer-ml/