Zero-shot image classification method, system, electronic device and storage medium

Abstract

The present disclosure provides a zero-sample image classification method, system, electronic device and storage medium. The method first extracts global features of an input image, and then learns the global features based on an attention mechanism to obtain multiple feature masks. calculating an adaptive threshold based on the maximum value of the plurality of maximum mask values ​​and a preset adaptive factor; obtaining a weighted global feature of the input image based on the adaptive threshold and the plurality of feature masks; Calculate the compatibility score of the weighted global feature and the semantic embedding vector of the unseen category, determine the maximum value of the plurality of compatibility scores as the highest compatibility score, and output the highest compatibility score with the highest compatibility The unseen category corresponding to the score is used as the category prediction result of the input image, so that through the threshold adaptive attention mechanism, the redundant features are suppressed while the robustness of the features is improved, and the classification accuracy is further improved. .

Description

technical field [0001] The present disclosure relates to the technical field of image classification, and in particular, to a zero-sample image classification method, system, electronic device and storage medium. Background technique [0002] Zero-shot learning is a type of small-shot learning. These concepts are inspired by human learning. Humans only need to learn a small number of examples to master a new concept, or even learn a new concept without examples. Babies can easily recognize an apple the next time they see a real apple by looking at an apple on a book. Students can also learn some new concepts or affairs based on the teacher's description. For example, after learning the description that a zebra is a horse with black and white stripes, students can easily recognize a zebra after seeing it. [0003] Zero-shot recognition models make predictions based on direct semantics, and these methods can be classified into direct attribute prediction (DAP) and indirect at...

AI Technical Summary

Problems solved by technology

The models established using the DAP and IAP methods have strong interpretability, but these two methods place attributes in an overly important position, and mislabeling of attributes will have a greater negative impact on the performance of such methods

Images