Feature extraction network training method and device, terminal equipment and medium

Abstract

The embodiment of the invention is suitable for the technical field of image processing, and provides a feature extraction network training method and device, terminal equipment and a medium, and the method comprises the steps: determining a background interval of each pixel point in a video frame sequence; according to the background interval, foreground pixel points in a target video frame are determined, and the target video frame is any video frame in the video frame sequence; acquiring a foreground image in each target video frame according to the foreground pixel points; determining a positive sample and a negative sample of a foreground image of the target video frame to construct a positive sample pair and a negative sample pair; and adopting the positive sample pair and the negative sample pair to carry out self-supervised training on a preset feature extraction network. The feature extraction network obtained by training through the method is high in matching degree in downstream tasks of target detection and segmentation.

Description

technical field [0001] The present application belongs to the technical field of image processing, and in particular relates to a training method, device, terminal equipment and medium of a feature extraction network. Background technique [0002] Self-supervised learning (self-supervised learning) mainly uses auxiliary tasks to mine its own supervision information from large-scale unsupervised data, and trains the network through the supervised information of this structure, so that it can learn valuable representations for downstream tasks. . [0003] Self-supervised learning methods have been widely used in the field of target detection, but the regions intercepted by existing self-supervised learning methods when sampling on the original image cannot reflect the complete information of the object, or contain a large amount of irrelevant information. Therefore, the currently learned features are incomplete or redundant, making the corresponding object detection poorly ma...

AI Technical Summary

Problems solved by technology

[0003] Self-supervised learning methods have been widely used in the field of target detection, but the intercepted areas of the existing self-supervised learning methods cannot reflect the complete information of the object when sampling on the original image, or contain a large amount of irrelevant information.

Therefore, the currently learned features are incomplete or redundant, making the corresponding object detection poorly matched with the downstream tasks of segmentation

Images