Recent work on curvilinear structure segmentation has mostly focused on backbone network design and loss engineering. The challenge of collecting labelled data, an expensive and labor intensive process, has been overlooked. While labelled data is expensive to obtain, unlabelled data is often readily available. In this work, we propose SemiCurv, a semi-supervised learning (SSL) framework for curvilinear structure segmentation that is able to utilize such unlabelled data to reduce the labelling burden. Our framework addresses two key challenges in formulating curvilinear segmentation in a semi-supervised manner. First, to fully exploit the power of consistency based SSL, we introduce a geometric transformation as strong data augmentation and then align segmentation predictions via a differentiable inverse transformation to enable the computation of pixel-wise consistency. Second, the traditional mean square error (MSE) on unlabelled data is prone to collapsed predictions and this issue exacerbates with severe class imbalance (significantly more background pixels). We propose a N-pair consistency loss to avoid trivial predictions on unlabelled data. We evaluate SemiCurv on six curvilinear segmentation datasets, and find that with no more than 5% of the labelled data, it achieves close to 95% of the performance relative to its fully supervised counterpart.

Building accurate regression models using limited data is a challenging problem in manufacturing data analysis. In this paper, we study a particular semisupervised learning problem where labeled data are limited, while unlabeled data are plentiful. In these conditions, conventional single-view learning methods are prone to overfitting. To tackle this problem, we develop a novel co-training technique, namely partial Bayesian co-training (PBCT). PBCT scales down the original set of features to create a partial view, and then exploit side information from the partial view to enhance the complete model. The PBCT model also allows integrating domain knowledge to enhance model accuracy. The proposed method is validated with experiments on industrial manufacturing data. The experimental results show that under a reduction of labeled data by up to 50%, a robust estimation is still attainable. This suggests that the PBCT model is a promising solution to a broad spectrum of applications.