Zhang, Wenze (China, Beihang University, School of Transportation Science and ), Lin, Chunmian (Beihang University), Tian, Daxin (Beihang University), Duan, Xuting (Beihang University), Zhou, Jianshan (Beihang University)

RailSAM: Taming SAM with Adapter for Railway Segmentation

Scheduled for presentation during the Video Session "On-Demand Video Presentations" (VP-VP), Saturday, November 22, 2025, 08:00−18:00, On-Demand Platform

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on April 2, 2026

Abstract

The emergence of vision foundation model, such as the Segment Anything Model (SAM), has brought groundbreaking advancements to downstream tasks due to its zero-shot transfer capability. In this work, RailSAM is proposed by incorporating the SAM generalist architecture with domain-specific adapter for railway segmentation. By leveraging the pretrained knowledge from the frozen SAM encoder, we seamlessly inject domain-specific information through visual prompts into the mask decoder via carefully designed adapters. On the publicly available RailSem19 dataset, our experiments demonstrate that RailSAM significantly outperforms existing task-specific methods and exhibits remarkable robustness under challenging conditions. We hope this work would inspire in-depth exploration of vision foundation model for intelligent railway transportation. The code will be open-source soon.