Using Deep Semantic Segmentation Model to Divide the Human Body Images

Deep semantic networks have achieved good results in object segmentation. Improvements have made to the deep network model to achieve human segmentation on a single CPU. In specific scenarios, the human body images can divided into multiple targets. Consider the spatial and temporal connectivity of different parts of the human body, and establish contextual connections using perspective clues. In addition to pixel level annotations, semantic regions should labeled with shapes or structures. Developed an advanced region detection method that integrates data with other pixel level segmentation data. Training and evaluation require a large number of data samples. We have built a human body dataset ourselves. This dataset comes from the internet and image collections. Label each human body based on 12 target categories. In order to quantitatively evaluate the segmentation effect, we conducted testing and statistics on all images in the dataset. Experiments have shown that deep semantic segmentation achieve good result in object segmentation.