A Few-Shot Semantic Segmentation Method Based on Feature Harmony Activation

Abstract

The invention discloses a small-sample semantic segmentation method based on harmonious activation of features, a small-sample semantic segmentation system, and a computer-readable storage medium. The method includes a process of training a segmentation model for semantic segmentation, and the segmentation model training process includes the following Steps: perform feature extraction on the support image and the query image; fuse the support feature and the query feature to obtain an intermediate feature activation map; update the intermediate feature activation map to obtain a feature harmony activation map; perform semantic segmentation on the feature harmony activation map, Obtain the segmentation map of the query image. The small-sample semantic segmentation method based on harmonious activation of features disclosed in the present invention can accurately and completely activate the target categories in the query features while fully retaining the detailed information in the support and query features.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a small-sample semantic segmentation method based on harmonious activation of features, which uses a small number of labeled support samples to perform semantic segmentation on unlabeled query samples. Background technique [0002] Deep learning has made tremendous progress in the fields of vision, text, speech, search, etc., largely thanks to a large number of annotated datasets. However, the labeling of datasets requires a lot of manpower and material resources, especially for semantic segmentation tasks, whose datasets require dense annotation at the pixel level, which is very expensive. Small sample learning can quickly learn and optimize the model by learning a small amount of labeled data, which can largely alleviate the problem of labeling costs. [0003] For small-sample semantic segmentation, the model is first subjected to feature extraction and lea...

AI Technical Summary

Problems solved by technology

[0005] However, the method of compressing the target category features into semantic vectors will lose a lot of detailed information. At the same time, comparing the semantic vectors with the query features will lead to insufficient interaction between the support and query feature elements, so that the target category features in the query features cannot be very accurate. Well activated, resulting in category misclassification and incomplete segmentation

Images