Small sample video target segmentation method based on dynamic prototype learning

Abstract

The invention discloses a small sample video target segmentation method based on dynamic prototype learning. The method comprises the following steps: acquiring a video target to be segmented; and processing a to-be-segmented video target by using the small sample video target segmentation model based on dynamic prototype learning to obtain a video target segmentation result. According to the small sample video target segmentation method provided by the invention, a dynamic prototype is adaptively learned by using an optimal transmission method, noise attention is effectively reduced, and a multi-level feature map is matched by adopting a guiding mode, so that the calculation amount is greatly reduced; according to the method, target information in a small amount of support set samples can be fully extracted, and the segmentation performance on a challenge set video is remarkably improved. The invention also discloses an electronic device, a storage medium and a computer program product for executing the small sample video target segmentation method based on dynamic prototype learning.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a training method of a small sample video target segmentation model and a video target segmentation method. Background technique [0002] Video object segmentation is a technique that predicts foreground object masks in each frame of a video, and has a wide range of applications in augmented reality, autonomous driving, video editing, and more. [0003] Video object segmentation in the prior art is usually based on semi-supervised and unsupervised. Among them, the semi-supervised method needs to give the target information of the first frame of each video, and then perform dense association of the target in the subsequent frames of the video. This process relies heavily on a large amount of densely segmented and annotated data, which is very time-consuming and labor-intensive. Unsupervised methods have low performance due to the lack of labeled data, which cannot meet the needs of...

AI Technical Summary

Problems solved by technology

Among them, the semi-supervised method needs to give the target information of the first frame of each video, and then carry out dense association of the target in the subsequent frames of the video. This process relies heavily on a large amount of densely segmented and labeled data, which is very time-consuming and labor-intensive; and Due to the lack of labeled data, the unsupervised method has low performance and cannot meet the needs of practical applications.

In addition, the above two methods cannot generalize well to new object categories, and the segmentation ability drops sharply on categories not seen in the training stage, which limits the scalability and practicality of video object recognition.

Images