Visual attribute recognition method and device and storage medium

Abstract

Embodiments of the present application provide a visual attribute recognition method and device and a medium. The method includes acquiring basic visual information of a target image by using a firstpart of a preset neural network model; using the second part of the preset neural network model to obtain the global visual information and the local visual information; according to the importance ofeach feature in the local vision information, processing the local vision information to obtain the first local adjustment information; according to the importance of each feature in the global vision information, processing the global vision information to obtain the first global adjustment information; adjusting the global vision information according to the first local adjustment information to obtain the adjusted global vision information; adjusting the local vision information according to the first global adjustment information to obtain the adjusted local vision information, and performing the visual attribute recognition on the target image according to the adjusted global vision information or / and the local vision information. The accuracy of visual attribute recognition can be improved.

Description

technical field [0001] The present application relates to the technical field of visual attribute recognition, and in particular to a visual attribute recognition method, device and storage medium. Background technique [0002] Visual attributes can be defined as mid-level semantic visual concepts, such as pointed nose and big eyes as facial attributes, and height and clothing style as pedestrian attributes. In recent years, visual attribute recognition has attracted more and more research interest, because the use of identified attributes can help advanced vision tasks such as zero-shot learning and person re-identification, or use these attributes alone for video surveillance and popular Apparel recommended. [0003] Many existing methods treat attribute recognition as a multi-task learning (MTL) problem and leverage deep neural networks (DNNs) to achieve state-of-the-art results, where each attribute recognition problem is considered as a task. DNN-based MTL methods usu...

AI Technical Summary

Problems solved by technology

A potential difficulty is that sharing hidden layers between unrelated tasks may harm the performance of a task by introducing knowledge that is not relevant to it. This is the so-called negative transfer problem.

Images