A zero-sample recognition method for fine-grained images

Abstract

The invention discloses a fine-grained image zero-sample identification method. The method is based on the zero-sample image classification technology of semantic decomposition and migration, which can fully mine the semantic information of two domains and generate unbiased semantic and visual expressions at the same time. At the same time, the method achieves state-of-the-art results on four public fine-grained classification datasets.

Description

technical field [0001] The invention relates to the technical field of image classification, in particular to a zero-sample identification method for fine-grained images. Background technique [0002] Traditional image classification techniques require massive data support and precise manual annotation. However, in the face of various task requirements, relying on manual annotation is an inefficient approach. [0003] In recent years, zero-shot image recognition techniques have received a lot of attention, with the aim of enabling models to recognize never-before-seen image categories. To achieve this task, different categories of unbiased semantic information are used to connect source domain data (visible data) and target domain data (unseen data) to make the model more robust to biases in different data domains. The usual practice is to map images and semantic attributes to the same embedding space, so that the corresponding images and semantic attributes can be correct...

AI Technical Summary

Problems solved by technology

Since the image of the target domain cannot be obtained during the training phase, the trained model is very sensitive to the deviation of the two domains, and the existing methods only use the semantic information of the target domain to obtain unbiased semantic or visual expressions, which are not sufficient. Make good use of semantic information

Images