Vehicle re-Identification

Introduction

Vehicle re-Identification aims to identify the samevehicle across the different cameras. It has useful applicationsin surveillance and intelligent transport systems. One of thefundamental challenges of vehicle re-identification is how tolearn robust and discriminative visual features given in smallinter-class similarity and large intra-class differences that don’tfollow the same distribution. In this project we propose to treatthe re-identification problem as a domain adaptation task. We implemented the Deep Joint Domain Adaptation algorithm totrain a model that could robustly detect the same classes evenwhen the images were obtained in different conditions.

Preparation

Dependencies

• Python 3
• PyTorch
• PIL
• Matplotlib
• POT (for optimal transport)

Dataset

• VeRi-776
• Download the data and save in ./data/VeRi folder

Report

You can find the report summarizing the results of the project here.

Authors

• Israfel Salazar
• Yujin Cho