Panoramic Image Segmentation on Individual Object Instances
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
We propose a novel task called panorama segmentation (PS), which unifies the traditionally separate domains of instance segmentation (identifying and segmenting individual object instances) and semantic segmentation (assigning class labels to each pixel). This research underscores the necessity of establishing a unified, detailed, and comprehensive framework for scene segmentation, representing a pivotal advancement for real-world vision systems. Previous attempts in computer vision to tackle related challenges in image and scene processing have struggled to gain traction, largely due to the absence of reliable evaluation metrics and the complexities associated with accurate identification. To address this issue, we propose a novel metric, panoramic quality (PQ), which offers a standardized and transparent approach to evaluating performance across all categories, encompassing both “stuff” and “things.” Leveraging this metric, we present an in-depth analysis of human and machine performance in PS across three widely used datasets, providing critical insights into this emerging task.
Fields of Study & Subjects
Advanced Neural Network Applications
Domain Adaptation and Few-Shot Learning
Visual Attention and Saliency Detection
Publication details
- Journal / source
- CRC Press eBooks
- Pages
- 51–55
- Publication date
- 31 March 2026
- Language
- EN
- Document type
- Book chapter