Person Re-ID Model Training Method Based on Heterogeneous Dual Network and Feature Consistency

Abstract

The invention discloses a domain adaptive pedestrian re-identification model training method, equipment, device, computer storage medium and pedestrian re-identification method based on heterogeneous dual networks and feature consistency. The invention designs a heterogeneous dual network framework, It contains two asymmetric branches, one of which uses convolution with limited receptive field to obtain local information, the other uses Transformer module to capture long-range dependencies, and uses the mutual learning of heterogeneous dual networks to improve the heterogeneity and complementarity between networks In order to reduce the interference of noise pseudo-labels in the network optimization process, a feature consistency loss is proposed, which does not need to rely on any label information, and pays more attention to the consistency of samples in the feature space In order to enhance the semantic information of the network, the present invention designs an adaptive channel mutual perception module to perform feature extraction on the salient area of ​​pedestrians, thereby improving the accuracy and efficiency of pedestrian re-identification.

Description

technical field [0001] The invention relates to the technical field of machine vision, in particular to a domain adaptive pedestrian re-identification model training method, equipment, device, computer storage medium and pedestrian re-identification method based on heterogeneous dual networks and feature consistency. Background technique [0002] Person re-identification is a very important research topic in the field of machine vision. Traditional person re-identification mainly uses a large amount of labeled image data for training in specific scenes. Although supervised learning methods have achieved good results, obtaining labeled data requires a lot of manpower and material resources. In addition, in practical applications, pedestrians have different appearances, backgrounds, and lighting conditions in different scenes, so that a model trained on one dataset cannot be directly applied to another dataset. The generalization of person re-id models to other domains is a r...

AI Technical Summary

Problems solved by technology

[0005] For this reason, the technical problem to be solved by the present invention is to overcome the problem that the training process in the prior art is seriously disturbed by pseudo-label noise

Images