IndoorMS: A Multispectral Dataset for Semantic Segmentation in Indoor Scene Understanding

Indoor scene understanding is an important task in computer vision. Typically, RGB information has been used as input for deep-learning-based semantic segmentation networks to achieve pixel-level scene understanding. However, indoor environments also offer rich information beyond the visible light spectrum, which existing research has largely overlooked. This study introduces IndoorMS, a dedicated multispectral dataset for semantic segmentation of indoor scenes. We collected images with a multispectral sensor across diverse indoor environment settings, including meeting rooms, halls, lounges, offices, corridors, and classrooms, in 17 buildings. This finely segmented dataset provides 19 semantic categories. Benchmark experiments are conducted using multiple representative semantic segmentation frameworks, followed by a detailed evaluation of their segmentation performance. This work is the first to support research on indoor multispectral scene understanding with a dedicated dataset. Key challenges associated with IndoorMS and future research directions are also highlighted. The project page for the IndoorMS dataset is available at https://zhuqinfeng1999.github.io/IndoorMS/. (The dataset will be publicly available for download after peer review.)