Panoptic Feature Pyramid Networks (Panoptic FPN) is a unified approach for instance and semantic segmentation. The method builds upon Mask R-CNN with a Feature Pyramid Network (FPN) backbone, adding a lightweight semantic segmentation branch. This approach enables the network to perform both instance and semantic segmentation simultaneously, achieving high accuracy and efficiency. The FPN backbone provides multi-scale features, which are used for both tasks. The semantic segmentation branch uses upsampling and aggregation of features from different FPN levels to generate dense-pixel outputs. The method is trained with a combined loss function that balances the instance and semantic segmentation losses. Experiments show that Panoptic FPN achieves competitive performance on COCO and Cityscapes datasets, with significant improvements over single-task models. The method is computationally efficient, requiring only a slight overhead over Mask R-CNN. It is also flexible, allowing the use of various backbones. Panoptic FPN demonstrates strong performance in panoptic segmentation, achieving results comparable to or better than separate models, with roughly half the computational cost. The method is a robust and effective baseline for panoptic segmentation, offering a simple and efficient solution for the joint task of instance and semantic segmentation.Panoptic Feature Pyramid Networks (Panoptic FPN) is a unified approach for instance and semantic segmentation. The method builds upon Mask R-CNN with a Feature Pyramid Network (FPN) backbone, adding a lightweight semantic segmentation branch. This approach enables the network to perform both instance and semantic segmentation simultaneously, achieving high accuracy and efficiency. The FPN backbone provides multi-scale features, which are used for both tasks. The semantic segmentation branch uses upsampling and aggregation of features from different FPN levels to generate dense-pixel outputs. The method is trained with a combined loss function that balances the instance and semantic segmentation losses. Experiments show that Panoptic FPN achieves competitive performance on COCO and Cityscapes datasets, with significant improvements over single-task models. The method is computationally efficient, requiring only a slight overhead over Mask R-CNN. It is also flexible, allowing the use of various backbones. Panoptic FPN demonstrates strong performance in panoptic segmentation, achieving results comparable to or better than separate models, with roughly half the computational cost. The method is a robust and effective baseline for panoptic segmentation, offering a simple and efficient solution for the joint task of instance and semantic segmentation.