Graph Convolutional Network for Scene Parsing

Semantic analysis of scene is one of the focuses in computer vision research, and scene parsing is a key technology to realize semantic segmentation and recognition of image. In some studies, researchers concentrated on image-level scene classification or attribute recognition , who integrated hand-craft image features and classification models, such as support vector machine, decision tree and multi-layer perceptron, to build scene representation and classification model. These methods sometimes were not robust for real-world applications, because the scene image contained considerable objects and structure information, which caused huge intra-class difference or inter-class similarity.