Method, system, electronic device and storage medium for training zero-shot classification model

Abstract

The invention provides a method, system, electronic equipment and storage medium for training a zero-sample classification model. The method includes performing multiple segmentation and reorganization on each sample image, and reorganizing the reorganized images after each segmentation and reorganization according to the mosaic parameters. Input the zero sample classification model in order from large to small, pass through the reference neural network layer and the fully convolutional neural network layer, obtain the weighted global features of the restructured image, and pass through the second neural network layer The network layer calculates the compatibility score of the weighted global feature and the semantic embedding vector of the unseen category, and obtains the predicted probability that the sample image belongs to the unseen category based on the compatibility score, so that the model can be small Gradually learn the local features of the target, and improve the classification accuracy of the zero-shot classification model.

Description

technical field [0001] The present invention relates to the technical field of zero-sample image classification, in particular to a method, system, electronic device and storage medium for training a zero-sample classification model. Background technique [0002] Zero-sample learning is a type of small-sample learning. These concepts are inspired by human learning. Humans can master a new concept with only a few examples, or even learn a new concept without examples. Babies can easily recognize that this is an apple by looking at the apple on the book, the next time they see a real apple. Students can also learn some new concepts or things 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 the zebra after seeing it. [0003] During zero-shot model training, once the neural network structure is determined, the scale of the input data has been determined, and all i...

AI Technical Summary

Problems solved by technology

However, these methods can only make the model pay attention to the global features, and cannot pay attention to the local features of each area of ​​each sample picture.

Images