A zero sample learning method based on a self-coding generative adversarial network

Abstract

The invention discloses a zero sample learning method based on a self-coding generative adversarial network. The method comprises the following steps: inputting visual features of visible category samples and corresponding category semantic features; Inputting a specific balance parameter Lambda and alpha numerical value; setting Initial values and learning rates of the parameters are set, using an Adam optimizer is to train the self-encoding confrontation generation network provided by the invention, and obtiaining model parameters of an encoder and a decoder; Inputting the semantic featuresof which the categories are not seen, and synthesizing the visual features of the corresponding categories by utilizing the trained model parameters; And classifying the test samples of which the categories are not seen. According to the invention, the semantic relationship between the visual modality and the category semantic modality can be effectively aligned. Visual information and category semantic information are fully fused together, semantic association between two modes can be more effectively mined, and more effective visual features are synthesized.

Description

technical field [0001] The invention relates to a zero-sample learning method. In particular, it concerns a zero-shot learning method based on autoencoder adversarial generative networks. Background technique [0002] Advances in deep learning techniques have greatly advanced the fields of machine learning and computer vision. However, most of these techniques are limited to supervised learning, which requires a large number of labeled samples to train the model. In reality, sample labeling is an extremely laborious task. Therefore, the lack of labeled samples is one of the bottlenecks affecting the development of current machine learning. A technology that can still identify these categories in the absence of visually labeled data of target categories is needed. Zero-shot learning is just such a type of technology. [0003] Zero-shot learning is a technique for identifying unseen categories (categories without training data) using data of visible categories, supplemented...

AI Technical Summary

Problems solved by technology

However, most of these models only focus on the semantic alignment relationship between category semantics and visual features and ignore the relationship between visual features and category semantics, which weakens the alignment relationship between different modalities.

Images