Video image segmentation method and device, storage medium, electronic equipment

Abstract

The present disclosure relates to a video image segmentation method and device, belonging to the technical field of video image segmentation. The method includes: using historical video images and mask images of the historical video images to perform machine learning to obtain an image segmentation model; The image segmentation model is adjusted by the first frame of the first frame, the previous frame of the frame to be segmented, and the mask image of the previous frame; the image to be segmented is segmented by using the adjusted image segmentation model. This method can improve the accuracy of video image segmentation.

Description

technical field [0001] The present disclosure relates to the technical field of video image segmentation, and in particular, to a video image segmentation method, a video image segmentation device, a computer-readable storage medium, and electronic equipment. Background technique [0002] With the rapid development of e-commerce, the competition among major e-commerce platforms is becoming more and more intense. Therefore, in order to improve their own competitiveness and provide users with more comprehensive product information, most e-commerce platforms record products through video recording and video segmentation to obtain views of all directions of the product so that users can get more Comprehensive product information. [0003] Video object segmentation can also be called VOS (Video Object Segmentation), which needs to extract video frames from the video frame by frame and segment the objects in the frame. In the existing video image segmentation method, the followi...

AI Technical Summary

Problems solved by technology

[0004] One is the Single-shot Video Object Segmentation (OSVOS, One-shot Video Object Segmentation) method, which fine-tunes the trained model through the first frame of the video image to be segmented, and then uses the fine-tuned model to generate subsequent frames. Image segmentation results; however, this method only uses the learning parameters of the first frame to fine-tune the model, and cannot predict large changes in subsequent frames, resulting in the splitting of subsequent frame segmentation;

[0005] The other is that the On-line Adaptive Video Object Segmentation (On-line Adaptation Video Object Segmentation) method is proposed. This method uses the results generated in the previous frame and the 0th frame to fine-tune the trained model for the following The generation of one frame is used as a guide. Compared with the segmentation method that only uses the first frame for guidance, this method improves the segmentation effect, but for problems such as object occlusion and object rotation, the segmentation result of the previous frame is not very good. Guiding the generation of the next frame image segmentation results, the segmentation results are getting worse and worse

Images