Active Learning for Imbalanced Civil Infrastructure Data

Aging civil infrastructures are closely monitored by engineers for damage and critical defects. As the manual inspection of such large structures is costly time-consuming, we working towards fully automating visual inspections to support prioritization maintenance activities. To that end combine recent advances in drone technology deep learning. Unfortunately, annotation costs incredibly high as our proprietary engineering dataset must be annotated highly trained engineers. Active learning is, therefore, a valuable tool optimize trade-off between model performance costs. Our use-case differs from classical active setting suffers heavy class imbalance consists much larger already labeled data pool than other research. We present novel method capable operating this challenging replacing traditional acquisition function with an auxiliary binary discriminator. experimentally show outperforms best-performing (BALD) 5% 38% accuracy on CIFAR-10 respectively.