Open-Vocabulary 3D Indoor Scene Segmentation: A Systematic Review
Abstract Open-vocabulary 3D indoor scene segmentation seeks to overcome the fixed-category limitation of traditional 3D segmentation tasks by transferring the rich semantic knowledge of large-scale vision-language models to the 3D space. This topic has gained traction with the emergence of vision-language models such as Contrastive Language-Image Pre-training (CLIP) model and continues to develop due to its fundamental importance across various domains in computer vision, robotics, and autonomous driving. Given the growing significance of this field and the rapid expansion of research on open-vocabulary 3D scene understanding, this systematic review examines 93 peer-reviewed papers to provide the first systematic analysis focused on open-vocabulary 3D indoor scene segmentation. These papers are identified and extracted from three databases (Scopus, IEEE Xplore, and ACM Digital Library) following multiple inclusion and exclusion criteria. We critically examine the studies while thematically grouping them based on their main representation: point cloud-based, NeRF-based, and 3DGS-based methods. Then, we synthesise the evaluation protocols adopted across the existing literature. Finally, we discuss the challenges and promising future directions to guide researchers toward the main research gaps in open-vocabulary 3D segmentation.
Authors
- Mohammed Bennamoun (ORCID: https://orcid.org/0000-0002-6603-3257)
- Farid Boussaïd (ORCID: https://orcid.org/0000-0001-7250-7407)
- Syed Mohammed Shamsul Islam (ORCID: https://orcid.org/0000-0002-3200-2903)
- Hussam Azzuni (ORCID: https://orcid.org/0000-0001-7267-8919)
- Viet H. Huynh (ORCID: https://orcid.org/0000-0003-1308-1164)
- Syed Afaq Ali Shah
Institutions
- Edith Cowan University (AU)
- The University of Western Australia (AU)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-17
- DOI
- https://doi.org/10.21203/rs.3.rs-11475551/v1
- Primary Topic
- 3D Shape Modeling and Analysis
- Type
- preprint