Grafted network for person re-identification

TL;DR

This paper introduces GraftedNet, a lightweight neural network for person re-identification that combines a ResNet-50 rootstock with a SqueezeNet scion, achieving high accuracy with fewer parameters.

Contribution

The paper proposes a novel grafted network architecture and an accompanying training method to reduce parameters and computation in person re-ID models.

Findings

Achieves over 93% Rank-1 accuracy on Market1501

Uses only 4.6 million parameters, significantly fewer than comparable models

Demonstrates effectiveness across three public benchmarks

Abstract

Convolutional neural networks have shown outstanding effectiveness in person re-identification (re-ID). However, the models always have large number of parameters and much computation for mobile application. In order to relieve this problem, we propose a novel grafted network (GraftedNet), which is designed by grafting a high-accuracy rootstock and a light-weighted scion. The rootstock is based on the former parts of ResNet-50 to provide a strong baseline, while the scion is a new designed module, composed of the latter parts of SqueezeNet, to compress the parameters. To extract more discriminative feature representation, a joint multi-level and part-based feature is proposed. In addition, to train GraftedNet efficiently, we propose an accompanying learning method, by adding an accompanying branch to train the model in training and removing it in testing for saving parameters and computation. On three public person re-ID benchmarks (Market1501, DukeMTMC-reID and CUHK03), the effectiveness of GraftedNet are evaluated and its components are analyzed. Experimental results show that the proposed GraftedNet achieves 93.02%, 85.3% and 76.2% in Rank-1 and 81.6%, 74.7% and 71.6% in mAP, with only 4.6M parameters.