Hotel-ID 2021

Overview

Human trafficking victims are sometimes photographed in hotels. Identifying these hotels is an important part of making a case against these victims’ traffickers -- for example, demonstrating that a victim was photographed in hotels in different states makes the offence a federal, rather than local or state crime. Performing this recognition task, however, is difficult for a variety of reasons. There are a huge number of possible hotel classes, with both high intraclass and low interclass variance -- rooms within a hotel may look quite different, while rooms from different instances of the same chain may look similar. Images typically only show a small fraction of the hotel room. To drive the development of new approaches in this difficult and important area of research, we are hosting HotelFinder, a hotel recognition challenge with data collected from the TraffickCam mobile application which allows every day travellers to submit images of their hotels to support investigations of human trafficking. In this contest, we will share more than 100,000 images from nearly 8000 hotels worldwide split between a training gallery with known hotel locations, and query images with unknown hotel identities. Through this competition, we challenge machine learning researchers and practitioners to develop algorithms that will help with an important step in understanding patterns of behavior in trafficking scenarios, and explore the ability of algorithms for real-world use cases with a large number of classes.

Competition

Start Date - 10th March 2021

End Date - 26th May 2021

Kaggle URL - https://www.kaggle.com/c/hotel-id-2021-fgvc8

Organizers

Abby Stylianou (Saint Louis University), Rashmi Kamath (Saint Louis University), Richard Souvenir (Temple University), Robert Pless (George Washington University)

Acknowledgements

Thank you to the users of the TraffickCam application for submitting images from their travels and to our collaborators at the Exchange Initiative and the National Center for Missing and Exploited Children. This work is funded by the National Institute of Justice (award #2018-75-CX-0038), and through a gift from Adobe Research.