Reproduce “PaveSAM Segment Anything for Pavement Distress” 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
Neema Jakisa Owor, Yaw Adu-Gyamfi, Armstrong Aboah, Mark Amo-Boateng
Researchers at the University of Missouri-Columbia and North Dakota State University developed PaveSAM, a fine-tuned Segment Anything Model, to accurately segment pavement distresses using bounding box prompts, significantly reducing the need for extensive pixel-level annotations. PaveSAM achieved a 34.9% Dice Similarity Coefficient improvement over the original SAM on their custom dataset and outperformed several state-of-the-art segmentation models.
Key insights
- Fine-tuning only the Mask Decoder of the general-purpose SAM can effectively adapt it for precise segmentation in specialized domains like pavement distress, saving computational resources.
- Bounding box prompts provide a practical and effective input mechanism for detailed pavement distress segmentation, offering a method to utilize datasets lacking pixel-level annotations.
- The effectiveness of zero-shot segmentation models in highly specialized visual domains benefits significantly from domain-specific adaptation to address unique feature characteristics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.