Multi-task network model, using method and device and storage medium

Abstract

The invention discloses a multi-task network model based on single excitation, a method and a device for simultaneously carrying out face detection, face key point positioning and face attribute analysis on an input image by using the model, and a storage medium. The model comprises a feature sharing module and a multi-task network module, and feature maps of different semantics can be shared through the feature sharing module, so that redundant calculation of features is reduced; secondly, a multi-level attention mechanism is introduced into each sub-task network, so that a feature channel and a feature region associated with a task can be enhanced; and thirdly, the plurality of sub-task networks connected in parallel receive the input image at the same time and perform parallel processing, so that the model is simpler and more efficient, face detection, key point positioning and attribute analysis can be performed at the same time on the premise of not executing a preposed task, andmulti-face attribute analysis can be directly performed on the input image. The method is widely applied to the technical field of image processing.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a multi-task network model based on a single excitation, and a method and device for simultaneously performing face detection, face key point location, and face attribute analysis on an input image using the model and storage media. Background technique [0002] Traditional face image analysis techniques are often only for a single task, such as age estimation, gender recognition, race classification, etc. For multi-attribute analysis of faces, it needs to be calculated in multiple times, which is very time-consuming and difficult to meet actual needs. In addition, the single-task face image analysis technology ignores the connection between various information and cannot make full use of the information contained in the face image. The facial features of the human face are different between different genders and different races. For example, there are differenc...

AI Technical Summary

Problems solved by technology

But, the shortcoming of its existence is: (1) the method described in this patent document does not comprise the human face detection step, so it is not applicable to the attribute analysis scene of many faces, in practical application, also needs an external human face detector to use (2) the multi-task CNN network model is a pipeline inference model, which includes a large number of cascade operations, which cannot be processed in parallel, which affects the inference speed; (3) the method design described in this patent document The three attribute analysis networks are independent, and they do not share network features, and they are optimized independently during the training process. This method will cause redundant feature calculations, which is not conducive to model expansion.

Images