In this paper, we propose ULOPS, an uncertainty-based open-set panorama segmentation framework for detecting unknown object instances. This framework leverages Dirichlet-based evidence learning to model prediction uncertainty and integrates semantic segmentation with uncertainty estimation, embeddings for prototype association, and a separate decoder for instance-centric prediction. Unknown instances are identified and segmented using uncertainty estimates during the inference process. To enhance the model's ability to distinguish between known and unknown objects, we introduce three uncertainty-based loss functions: Uniform Evidence Loss, Adaptive Uncertainty Separation Loss, and Contrastive Uncertainty Loss. We evaluate the open-set performance by extending the KITTI-360 benchmark and introducing a new open-set evaluation on nuScenes, experimentally demonstrating that our framework outperforms existing methods.
