Method for weakening influence of resampling on pedestrian re-identification

Abstract

The invention provides a method for weakening the influence of resampling on pedestrian re-identification, and belongs to the field of computer vision. The method comprises the steps: preprocessing a training data set, and calculating the sampling probability of each picture through a resampling strategy; constructing a feature extraction network, and extracting feature representation of the input pedestrian picture through the feature extraction network; calculating a loss function of the feature extraction network, and adjusting a weight ratio of inter-class loss to intra-class loss in the loss function to enable the loss function of the feature extraction network to be minimum; and iteratively training the pedestrian re-identification model until the pedestrian re-identification model reaches a convergence state. According to the method, the influence caused by the fact that feature points of pedestrian re-identification are more dispersed in a feature space can be reduced by weakening long-tail distribution and re-sampling.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a method for reducing the influence of resampling on pedestrian re-identification. Background technique [0002] Pedestrian re-identification, as a pedestrian recognition method that assists face recognition, can give full play to excellent results in situations where pedestrians are sideways, facing away, and with occluders on their faces. Therefore, pedestrian re-identification has received more and more attention in academia and industry more and more people's attention. The general method of pedestrian re-identification is to design a reasonable algorithm to make the algorithm converge and achieve a good enough recognition and classification effect by using the training data set. However, most of the existing data sets have long-tailed distributions. [0003] The long-tail distribution means that the number of pictures in each category is different, and the proportion may var...

AI Technical Summary

Problems solved by technology

This will bring about a problem. The sampling probability of the pedestrian pictures at the tail is relatively high, which will change the distribution of feature space data and affect the robustness of the model.

Images