Interactive video segmentation method under machine vision

The invention relates to a video segmentation technology in the field of robots, and particularly discloses an interactive video segmentation method under machine vision. The interactive video segmentation method under machine vision comprises the following steps of enabling a user to demonstrate object movement and different transformations on line, and teaching a label, calculating an optical flow of the input image by using SimpleFlow, building a teacher model, building a student model, and finely adjusting the teacher-student model. In a training stage, a user can demonstrate object movement and different transformations and generate a pseudo label for the object. In the reasoning process, objects can be segmented without manually segmenting annotations, and the speed is higher. 本发明涉及机器人领域视频分割技术，特别公开了一种机器视觉下的交互式视频分割方法。该机器视觉下的交互式视频分割方法，包括如下步骤：用户在线演示对象移动与不同的转换，并传授标签；使用SimpleFlow计算输入图像的光流；搭建教师模型；搭建学生模型；微调教师-学生模型。本发明在训练阶段，用户能够演示对象移动与不同的转换，并为其生成一个伪标签。在推理过程中可以在不需要手动分割注释的情况下分割对象，且速度更快。