The paper introduces the Bilateral Segmentation Network (BiSeNet), a novel approach for real-time semantic segmentation that balances speed and accuracy. BiSeNet consists of two main components: the Spatial Path and the Context Path. The Spatial Path, designed to preserve spatial information, uses a small stride to generate high-resolution features. The Context Path, aimed at providing a large receptive field, employs a fast downsampling strategy. These two paths are combined using a Feature Fusion Module to efficiently integrate features. The architecture is evaluated on Cityscapes, CamVid, and COCO-Stuff datasets, achieving a 68.4% Mean IOU on Cityscapes with a speed of 105 FPS on an NVIDIA Titan XP card, outperforming existing methods with comparable performance. The paper also discusses related work and provides detailed experimental results, demonstrating the effectiveness of BiSeNet in both speed and accuracy.The paper introduces the Bilateral Segmentation Network (BiSeNet), a novel approach for real-time semantic segmentation that balances speed and accuracy. BiSeNet consists of two main components: the Spatial Path and the Context Path. The Spatial Path, designed to preserve spatial information, uses a small stride to generate high-resolution features. The Context Path, aimed at providing a large receptive field, employs a fast downsampling strategy. These two paths are combined using a Feature Fusion Module to efficiently integrate features. The architecture is evaluated on Cityscapes, CamVid, and COCO-Stuff datasets, achieving a 68.4% Mean IOU on Cityscapes with a speed of 105 FPS on an NVIDIA Titan XP card, outperforming existing methods with comparable performance. The paper also discusses related work and provides detailed experimental results, demonstrating the effectiveness of BiSeNet in both speed and accuracy.