Answers are grounded in this paper and cite their sources.
This study investigates the feasibility of a fine-tuned Segment Anything Model (SAM) for forest road segmentation, utilizing high-resolution imagery from the Compact Advanced Satellite 500-1 (CAS500-1) as a case study.Due to the lack of labeled or domain-specific datasets for forest roads, a custom training dataset was constructed from Google Maps (GM) imagery, enabling the fine-tuning of SAM with a Vision Transformer-Huge (ViT-H) backbone.The results were evaluated against manually labeled ground truth, OpenStreetMap (OSM), and the Forest Big Data Exchange Platform (FBDEP) in Korea.Evaluation metrics included mean Intersection over Union (mIoU) and area-weighted mIoU (AWmIoU).The finetuned model achieved notable improvements compared to the baseline SAM (mIoU: 0.6190 to 0.7786; AWmIoU: 0.6970 to 0.8369).These findings confirm the value of domain-specific fine-tuning and demonstrate the potential of SAM for extracting forest roads.
No takes yet. Share an insight, caveat, or question.
Lee et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: