Uncertainty Aware Panoptic Segmentation

In the field of computer vision, Uncertainty-Aware Panoptic Segmentation is an emerging task that introduces three levels of difficulty labels in pixel annotations, allowing the model to reasonably compensate for errors in difficult regions during prediction. This task aims to enhance the robustness and accuracy of models in complex scenes by incorporating the Uncertainty-Aware Panoptic Quality (UPQ) evaluation metric, which further optimizes the traditional panoptic quality assessment system.