Search
Close this search box.
  • About Us
    • Governance
    • Our Team
    • Labs
    • Open positions
    • Alumni
  • Research
    • Wearable robotics
      • ReGait++
      • Third Arm
    • Rescue Robotics
      • Flying Robots
      • Legged robots
      • Collaboration, Learning and Tests
    • Educational robotics
      • Cellulo
      • Thymio
    • Publications
    • Open Science
  • Education
    • How to get into robotics
      • How our professors got into robotics
      • How our researchers got started
    • Robotics for kids
      • Activities for Kids
      • R2t2 Mars Mission
    • Graduate Programs in Robotics
    • Resources
  • Tech Transfer
    • Researchers expertise
    • Spin-Offs
    • Swiss Robotics Day
    • Swiss Robotics Ecosystem
    • Resources
  • Equal Opportunities
    • Robotics for Girls
    • Master students
    • Young Researchers
    • Building Equality
    • Women in Robotics
    • Resources
  • Awards & Programs
    • Spin-Fund
    • Master Students Exchange Program
    • PhD/Postdoc exchange program
    • Award for Career Development
    • Swiss Robotics Masters Award
    • Award for Scientific Visibility
    • Women keynote speakers
    • Best Paper Award
    • Ukrainian researchers
  • Outreach
    • Swiss Drone Days
    • Cybathlon
    • Swiss Robotics Days
    • Other Outreach Events
  • Contact
Menu
  • About Us
    • Governance
    • Our Team
    • Labs
    • Open positions
    • Alumni
  • Research
    • Wearable robotics
      • ReGait++
      • Third Arm
    • Rescue Robotics
      • Flying Robots
      • Legged robots
      • Collaboration, Learning and Tests
    • Educational robotics
      • Cellulo
      • Thymio
    • Publications
    • Open Science
  • Education
    • How to get into robotics
      • How our professors got into robotics
      • How our researchers got started
    • Robotics for kids
      • Activities for Kids
      • R2t2 Mars Mission
    • Graduate Programs in Robotics
    • Resources
  • Tech Transfer
    • Researchers expertise
    • Spin-Offs
    • Swiss Robotics Day
    • Swiss Robotics Ecosystem
    • Resources
  • Equal Opportunities
    • Robotics for Girls
    • Master students
    • Young Researchers
    • Building Equality
    • Women in Robotics
    • Resources
  • Awards & Programs
    • Spin-Fund
    • Master Students Exchange Program
    • PhD/Postdoc exchange program
    • Award for Career Development
    • Swiss Robotics Masters Award
    • Award for Scientific Visibility
    • Women keynote speakers
    • Best Paper Award
    • Ukrainian researchers
  • Outreach
    • Swiss Drone Days
    • Cybathlon
    • Swiss Robotics Days
    • Other Outreach Events
  • Contact
Search
Close this search box.
  • NCCR Robotics
    • Highlights
    • Governance
    • Our team
    • Labs
    • Past programs
    • Alumni
    • Open positions in robotics
  • Research
    • Wearable Robotics
      • ReGait++
      • Third Arm
    • Rescue Robotics
      • Flying robots
      • Legged robots
      • Collaborations, learning and tests
    • Educational Robotics
      • Cellulo
      • Thymio
    • Publications
    • Open Science
    • Research resources
  • Education
    • How to get into robotics
      • How our professors got into robotics
      • How our researchers got started
    • Robotics for kids
      • Activities for kids
      • R2t2 Mars mission
    • Graduate programs in robotics
    • Resources
  • Tech Transfer
    • Researchers expertise
    • Spin-Offs
    • Swiss Robotics Day
    • Swiss Robotics Ecosystem
    • Resources
  • Equal Opportunities
    • Building equality
    • Women in Robotics
    • Resources
  • Outreach
    • Swiss Robotics Days
    • Swiss Drone Days
    • Cybathlon
    • Photo exhibit
    • Other Outreach Events
Menu
  • NCCR Robotics
    • Highlights
    • Governance
    • Our team
    • Labs
    • Past programs
    • Alumni
    • Open positions in robotics
  • Research
    • Wearable Robotics
      • ReGait++
      • Third Arm
    • Rescue Robotics
      • Flying robots
      • Legged robots
      • Collaborations, learning and tests
    • Educational Robotics
      • Cellulo
      • Thymio
    • Publications
    • Open Science
    • Research resources
  • Education
    • How to get into robotics
      • How our professors got into robotics
      • How our researchers got started
    • Robotics for kids
      • Activities for kids
      • R2t2 Mars mission
    • Graduate programs in robotics
    • Resources
  • Tech Transfer
    • Researchers expertise
    • Spin-Offs
    • Swiss Robotics Day
    • Swiss Robotics Ecosystem
    • Resources
  • Equal Opportunities
    • Building equality
    • Women in Robotics
    • Resources
  • Outreach
    • Swiss Robotics Days
    • Swiss Drone Days
    • Cybathlon
    • Photo exhibit
    • Other Outreach Events

Deep visual Re-identification with confidence

Publication date:

  • May 2021

Authors:

George Adaimi
Sven Kreiss
Alexandre Alahi

Transportation systems often rely on understanding the flow of vehicles or pedestrian. From traffic monitoring at the city scale, to commuters in train terminals, recent progress in sensing technology make it possible to use cameras to better understand the demand, i.e., better track moving agents (e.g., vehicles and pedestrians). Whether the cameras are mounted on drones, vehicles, or fixed in the built environments, they inevitably remain scatter. We need to develop the technology to re-identify the same agents across images captured from non-overlapping field-of-views, referred to as the visual re-identification task. State-of-the-art methods learn a neural network based representation trained with the cross-entropy loss function. We argue that such loss function is not suited for the visual re-identification task hence propose to model confidence in the representation learning framework. We show the impact of our confidence-based learning framework with three methods: label smoothing, confidence penalty, and deep variational information bottleneck. They all show a boost in performance validating our claim. Our contribution is generic to any agent of interest, i.e., vehicles or pedestrians, and outperform highly specialized state-of-the-art methods across 6 datasets. The source code and models are shared towards an open science mission.

Transportation systems often rely on understanding the flow of vehicles or pedestrian. From traffic monitoring at the city scale, to commuters in train terminals, recent progress in sensing technology make it possible to use cameras to better understand the demand, i.e., better track moving agents (e.g., vehicles and pedestrians). Whether the cameras are mounted on drones, vehicles, or fixed in the built environments, they inevitably remain scatter. We need to develop the technology to re-identify the same agents across images captured from non-overlapping field-of-views, referred to as the visual re-identification task. State-of-the-art methods learn a neural network based representation trained with the cross-entropy loss function. We argue that such loss function is not suited for the visual re-identification task hence propose to model confidence in the representation learning framework. We show the impact of our confidence-based learning framework with three methods: label smoothing, confidence penalty, and deep variational information bottleneck. They all show a boost in performance validating our claim. Our contribution is generic to any agent of interest, i.e., vehicles or pedestrians, and outperform highly specialized state-of-the-art methods across 6 datasets. The source code and models are shared towards an open science mission.

Reference
Doi: 10.1016/j.trc.2021.103067
Published in: Transportation Research Part C: Emerging Technologies,
 Volume 126,
Read the paper
Version: Published
Access: Open Access
Download
  • Innovation Booster Robotics
  • News
  • Past Events
  • Contact
  • Innovation Booster Robotics
  • News
  • Past Events
  • Contact
  • Journalists and media
  • News bulletin archive
  • Newsletter archive
  • Journalists and media
  • News bulletin archive
  • Newsletter archive
Twitter Linkedin Youtube
NCCR Robotics partner institutions

Leading house

epfl_white

Co-Leading house

eth_white
unizuerich_white
idsia_white
unibe_all_white
empa-all-white
unibasel_white

This website was built by NCCR Robotics and provides useful information on Swiss Robotics. From December 2022, it is no longer updated.

Copyright © 2022 NCCR Robotics. All rights reserved.