Reproduce “Self-Supervised Multi-Scale Transformer with Attention-Guided Fusion for Efficient Crack Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Blessing Agyei Kyem, Joshua Kofi Asamoah, Eugene Denteh, Andrews Danyo, Armstrong Aboah
Researchers at North Dakota State University developed Crack-Segmenter, a fully self-supervised framework for pixel-level pavement crack segmentation that operates without any manual annotations during training. This approach achieved superior performance compared to thirteen state-of-the-art supervised methods on ten public crack datasets, with, for example, an mIoU of 0.8875 on the CFD dataset.
Key insights
- It is feasible to achieve effective pixel-level pavement crack segmentation entirely without manual annotations, surpassing the performance of many supervised approaches.
- Combining multi-scale feature embedding, directional attention, and adaptive fusion within a self-supervised framework effectively captures complex crack patterns and suppresses noise.
- Cross-scale consistency losses can generate sufficient supervisory signals from unlabeled data to train deep learning models for accurate semantic segmentation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.