Vehicle re-identification model compression method and system based on pruning and lightweight convolution

Abstract

The embodiment of the invention discloses a vehicle re-identification model compression method and system based on pruning and lightweight convolution, and the method comprises the steps: carrying out the pre-training of a backbone network of a to-be-compressed vehicle re-identification model, carrying out the pruning of the pre-trained backbone network, recovering the precision through the re-training, carrying out lightweight convolution design on a feature pyramid module in the vehicle re-identification model, combining a compact backbone network with the lightweight feature pyramid module, extracting features from the backbone network, and carrying out feature fusion on the feature pyramid module to obtain a lightweight vehicle re-identification model based on feature pyramid joint representation. According to the method, a complex and high-performance vehicle re-identification model is used as an input model, a convolution kernel with relatively low importance in a backbone network is automatically selected and pruned, and a convolution mode in a feature pyramid module is improved, so that the parameter quantity and the calculation quantity are effectively reduced, and a relatively compact model with equivalent precision is generated.

Description

technical field [0001] The embodiment of the present invention relates to the technical field of machine learning, and in particular to a method and system for compressing a vehicle re-identification model based on pruning and lightweight convolution. Background technique [0002] In recent years, with the rapid improvement of graphics processing unit (GPU) performance, deep neural network (DNN) has also made great development achievements with the support of powerful computing resources, and has repeatedly achieved success in many visual recognition tasks. However, because the mainstream deep learning network model has defects such as high computational complexity, large memory usage, and long time consumption, it is difficult to deploy in mobile devices with limited computing resources or applications with strict latency requirements, such as real-world scenarios. Vehicle re-identification task. Model compression refers to obtaining a more compact network by performing li...

AI Technical Summary

Problems solved by technology

[0003] For this reason, embodiments of the present invention provide a method and system for compressing vehicle re-identification models based on pruning and lightweight convolution to solve the problems of high computational complexity, large memory usage, and long time-consuming problems of existing vehicle re-identification models.

Images