Fast video portrait segmentation method and device based on FastPortrait model and medium

The invention discloses a fast video portrait segmentation method and device based on a FastPortrait model and a medium, and the method comprises the steps: decomposing a video to be subjected to image segmentation into a plurality of continuous frames of images, and carrying out the portrait segmentation through a portrait segmentation model trained in advance; the portrait segmentation model adopts a FastPortrait model of an encoder-decoder structure, and a memory matching module is arranged between the lowest layers of an encoder and a decoder; and the memory matching module is used for extracting and enhancing portrait features in the feature image output by the encoder according to the historical image feature memory bank. According to the method, the portrait in the video can be quickly and accurately predicted and segmented. 本发明公开了一种基于FastPortrait模型的视频人像快速分割方法、设备和介质，方法为：将待分像分割视频分解为连续多帧图像，使用预先训练好的人像分割模型进行人像分割；所述人像分割模型采用编码器-解码器结构的FastPortrait模型，编码器与解码器在最低层之间设置有记忆匹配模块；所述记忆匹配模块，根据历史图像特征记忆库，对编码器输出的特征图像中的人像特