Double-branch cross-domain pedestrian re-identification method based on attention calibration

Abstract

The invention discloses a double-branch cross-domain pedestrian re-identification method based on attention calibration, and the method comprises the steps: improving a ResNet-50 reference network through an attention calibration module, introducing a domain invariant feature supplement branch and a domain specific feature global branch, carrying out the linear weighting of domain invariant feature information and domain specific feature information, and obtaining a fusion feature, and the fusion features are normalized to obtain final features of the model. The reference network can fully extract channel information and space information of the features, and the problem of feature misalignment is solved. Model parameter quantity can be reduced by introducing domain specific feature global branches, training deviation is corrected in combination with a reference network, and meanwhile model complexity is reduced, so that the model can fully extract domain specific features. By introducing domain invariant feature supplementary branches, source domain complementary information obtained from calibration features of a specific domain is enriched, and domain specific features extracted from domain specific feature global branches are combined, so that the model has high generalization ability.

Description

technical field [0001] The invention relates to a pedestrian re-identification method, in particular to a double-branch cross-domain pedestrian re-identification method based on attention calibration. Background technique [0002] As one of the most attractive research tasks in the computer vision community, person re-identification aims to judge whether the same pedestrian is contained in views from non-overlapping cameras. It plays an important role in various surveillance applications, such as pedestrian retrieval and public safety incident detection, so it has received the attention of many scholars in the past few years, however, the person re-identification model is applied to new fields after being trained in the source domain The cross-domain recognition problem often occurs when the performance of the dataset is significantly degraded. Most of the existing researches focus on fully supervised person re-identification and have made encouraging progress. However, the...

AI Technical Summary

Problems solved by technology

He et al. used knowledge distillation to train multi-source domain data through the student model and the teacher model, and transferred the knowledge learned by the teacher model to the student model with relatively weak learning ability, so as to enhance the generalization ability of the student model. These methods tends to be overly complex and only removes style differences from a given source domain, so it lacks the ability to sufficiently attenuate new styles for unseen domains

Images