Image recognition training method and device, computer equipment and storage medium

Abstract

The invention discloses an image recognition training method and device, computer equipment and a storage medium, and the method comprises the steps: constructing a corresponding extension attribute based on an entity attribute of an entity type; based on the entity attribute and the extension attribute of the entity type, performing semantic consistency synthesis on the known visual features of the entity type, and obtaining known synthesis features corresponding to the entity attribute and the extension attribute respectively; and taking the known class visual features, the known class synthetic features, the entity attributes and the extension attributes as input data of a to-be-trained classification model, taking the class prototypes and the entity types corresponding to the known class visual features as output data of the to-be-trained classification model, training the to-be-trained classification model, and finally generating a trained classification model. According to the method, the consistency of visual features and semantic features of entity types can be effectively kept, and the accuracy of zero sample recognition is improved while the training stability is guaranteed.

Description

technical field [0001] The present invention relates to the technical field of image classification, and in particular, to an image recognition training method, device, computer equipment and storage medium. Background technique [0002] Traditional image classification can only classify the class samples that appear in the training set, but cannot classify the class samples that do not appear in the training set. However, in practical applications, new categories are constantly emerging, so it is necessary to collect a large amount of labeled data of new categories and retrain the classifier, which is time-consuming and labor-intensive, and in some fields, it is not even possible to obtain a sufficient amount of labeled data for new categories. In response to this problem, Zero-Shot Learning (ZSL) came into being. [0003] The purpose of zero-shot learning is to use only the visual features and semantic representations of visible class samples to train the model, and then ...

AI Technical Summary

Problems solved by technology

[0005] Embodiments of the present invention provide an image recognition training method, device, computer equipment, and storage medium to solve the problem of lack of diversity and differentiation in visual features of synthetic models that recognize unknown types

Images