Generalized zero-sample target classification method and device based on external distribution sample detection and related equipment

Abstract

The invention discloses a generalized zero-sample target classification method and device based on external distribution sample detection and related equipment. According to the method, an external distribution sample detector is trained by utilizing data of known classes and corresponding class semantic attributes, and each class is expressed as von Mises-Fisher (vMF) distribution in an implicitspace, so that a flow pattern boundary of each class is obtained. According to the flow pattern boundary of the known class, the provided external distribution sample detector can distinguish the characteristics of the unknown class from the characteristics of the known class. Therefore, the generalized zero sample classification problem can be simplified into a supervised classification problem and a traditional zero sample target classification problem, the feature confusion problem and the deviation problem in the generalized zero sample classification problem are avoided, and therefore thegeneralized zero sample classification performance is greatly improved. The method can be applied to an application environment which lacks training data and needs to identify unknown samples, such as an intelligent robot system, an intelligent recommendation system, a social media information filtering system and the like.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a generalized zero-sample target classification method, device and related equipment based on out-of-distribution sample detection. Background technique [0002] Generalized zero-shot classification is an important task in computer vision, and has broad application scenarios in tasks such as intelligent robots, intelligent recommendation, and social media information filtering. Previous generalized zero-shot recognition algorithms can be divided into two categories: embedding-based methods and synthetic feature-based methods. The main purpose of embedding-based methods is to establish a mapping between visual space and semantic space so as to measure the similarity of visual features and semantic attributes in the same space. The disadvantage of this method is that it is usually affected by the bias problem, that is, the features of the unknown class will be projected ...

AI Technical Summary

Problems solved by technology

[0003] However, this binary classifier is difficult to train due to the lack of training data for unknown classes

Images