This repository implements a CVPR 2020 research paper on semantic segmentation and scene parsing. The core contribution is a strip pooling technique that rethinks spatial pooling by using long narrow kernels instead of traditional square pooling windows. The implementation builds on semseg and PyTorch-Encoding libraries, supports synchronized batch normalization across multiple GPUs, and provides pretrained ResNet50/101 backbones. The model is designed for dense pixel-wise prediction tasks likeADE20K scene parsing.