Multi-task learning model construction and optimization method based on deformable convolution

Abstract

The invention discloses a multi-task learning model construction and optimization method based on deformable convolution, and belongs to the technical field of deep learning, and the method comprisesthe steps: introducing the deformable convolution into multi-task learning during the design of a network structure of a multi-task learning model, and sequentially connecting a deformable convolutionlayer and a feature alignment layer to construct a specific task deformable module; directly connecting the specific task deformable module with the feature sharing block, performing adaptive adjustment according to the content features of the sub-tasks, and performing differentiated specific task feature extraction and feature fusion to form a whole multi-task learning network framework; in theaspect of multi-task optimization design, utilizing the weight zero setting operation to achieve the dynamic balance of sub-task optimization. By means of the method, the performance of the multi-tasklearning model is remarkably improved.

Description

technical field [0001] The invention belongs to the technical field of deep learning, and relates to a multi-task learning network model through deformable convolution in deep learning, in particular to a method for building and optimizing a multi-task learning model based on deformable convolution. Background technique [0002] Most of the current network models are based on single-task design, that is, one network model only targets one specific task. However, in the real world, visual tasks are related to each other. The single-task network model isolates real problems from each other, ignores the rich correlation information between problems, and hinders further performance improvement. The multi-task network trains and learns multiple related tasks together, explores and mines the rich associated information contained in multiple related tasks, and helps improve the generalization performance of all related tasks. These characteristics make multi-task learning graduall...

AI Technical Summary

Problems solved by technology

[0008] Aiming at the problems existing in the above-mentioned prior art, the present invention provides a multi-task learning model construction and optimization method based on deformable convolution, which solves the following two technical problems: Due to the limitation, it is impossible to extract more distinguishable specific task features. The present invention introduces deformable convolution in the field of multi-task learning to build a specific task deformable module. Each module includes two parts: a deformable convolution layer and a feature alignment layer.

Images