ACM Multimedia 2020 Tutorial on

Effective and Efficient: Toward Open-world Instance Re-identification


Instance re-identification (re-ID) is a multimedia technology that finds a certain person/vehicle/object of interest in a large amount of videos. It facilitates various applications that require painful and boring video watching, including searching for video shots related to an actor of interest from TV series, a lost child in a shopping mall from camera videos, a suspect vehicle from a city surveillance system. Its efficiency and effectiveness accelerate the process of video analysis. In recent years, existing technologies have only been evaluated on standard benchmarks. Although we have made significant advances in standard datasets, it is still far away to design an open-world re-identification system.

In this tutorial, we summarize re-ID technologies and provide an overview. We'll introduce fundamental technologies, existing challenges, trends, etc. This tutorial would be useful for the multimedia content analysis and system-level multimedia retrieval, especially for an effective and efficient open-world re-ID system for the practical, large-scale, and open-set domain.

  • First, we will conduct a brief review for general person re-ID, where the person's appearance variation, the short-term environment change and the intra-modality discrepancy work as the main challenge. We will introduce new trends of person re-ID system that more practical in open-world conditions, consisting of group, long-term, and cross modality. Representative approaches, comparisons and discussions will be given.
  • Second, vehicle re-ID aims to instantly discover, locate, and track the target vehicles in large-scale urban surveillance system. We will survey the recent published massive vehicle re-ID datasets and deep learning vehicle re-ID methods, describe critical future directions in progressive vehicle re-ID with multi-modality information, and brief some critical yet under-developed issues.
  • Third, we will review state-of-the-art algorithms of approximate nearest neighbor search system. The design of the search algorithm is critical for performance. We will provide a practical guide to select the best algorithm for a given application system.
  • Finally, we will make a brief summary of instance re-ID, and show some trends of this task. In particular, we will introduce the INstance Search (INS) task and our effort and insights on this task.


  • An opening of the tutorial slides video
  • New trends of person re-ID system - Zheng Wang slides video
  • Vehicle re-ID: past, present and future - Wu Liu slides video
  • Billion scale approximate nearest neighbor search - Yusuke Matsui slides video
  • Is instance search a solved problem? - Shin'ichi Satoh slides video
  • QA Time

  • Date: 2020.10.16 The sessions are over. If you have any questions, please drop an email to us. We appreciate your interest in this topic and expect collaboration of any kind.
  • Program Zoom New York | UTC-4 Beijing | UTC +8 Tokyo | UTC +9 Seattle | UTC -7 UTC
    QA Link 9:00-10:00 21:00-22:00 22:00-23:00 6:00-7:00 13:00-14:00
    QA - Mirrored Link 19:00-20:00 7:00-8:00 +1 Day 8:00-9:00 +1 Day 16:00-17:00 23:00-24:00


    Zheng Wang

    National Institute of Informatics

    Wu Liu

    AI Research of

    Yusuke Matsui

    The University of Tokyo

    Shin'ichi Satoh

    National Institute of Informatics