The paper introduces the Pyramid Scene Parsing Network (PSPNet), a novel framework for scene parsing that addresses the challenges of unrestricted open vocabulary and diverse scenes. PSPNet leverages global context information through a pyramid pooling module, which aggregates context information from different regions to enhance pixel-level predictions. The proposed method effectively captures global prior representations, leading to superior performance on various datasets, including the ImageNet scene parsing challenge 2016, PASCAL VOC 2012, and Cityscapes. PSPNet achieves state-of-the-art results, outperforming existing methods in terms of mean intersection over union (mIoU) and accuracy. The paper also discusses the optimization strategy using deeply supervised loss and provides detailed implementation details, making the code and trained models publicly available.The paper introduces the Pyramid Scene Parsing Network (PSPNet), a novel framework for scene parsing that addresses the challenges of unrestricted open vocabulary and diverse scenes. PSPNet leverages global context information through a pyramid pooling module, which aggregates context information from different regions to enhance pixel-level predictions. The proposed method effectively captures global prior representations, leading to superior performance on various datasets, including the ImageNet scene parsing challenge 2016, PASCAL VOC 2012, and Cityscapes. PSPNet achieves state-of-the-art results, outperforming existing methods in terms of mean intersection over union (mIoU) and accuracy. The paper also discusses the optimization strategy using deeply supervised loss and provides detailed implementation details, making the code and trained models publicly available.