ABOUT THE CHALLENGE

Research on 3D mapping plays a vital role in improving our quality of life through numerous smart city applications. However, acquiring 3D data can be expensive, rendering frequent or large-scale data updates impractical. To address this issue, AI technologies are being explored to infer 3D information from easily accessible 2D images.

This challenge aims to address the problem statement of “How can AI be used to accurately extract camera pose data from 2D images?” where camera pose is a term used in computer vision (CV) to describe the position and orientation of a camera in 3D space relative to reference coordinate systems.

By leveraging advancements in visual odometry, we hope to push the boundaries of 3D learning and make it more affordable and sustainable for a broader range of real-world applications.

CHALLENGE TASK

The task requires participants to develop AI models that can accurately extract camera pose data from a given set of 2D monocular images and submit the results in given plain text format.

CHALLENGE STRUCTURE & TIMELINE

Please click here to enter the AISG-SLA Visual Localisation Challenge Competition Platform page.

Start Date: 26 May 2023
Entry Deadline: 26 July 2023 (AoE)
Winner Announcement: 14 – 16 August 2023
Final Presentation and Award Ceremony: 19 – 25 August 2023 (hosted by IJCAI-23)

The challenge will take the form of an online AI competition, where participants will be given access to a large training dataset to develop their AI models. They will then submit their camera pose predictions (in plain text format) against a given test dataset. Performance will be measured by translation and rotation errors on the test dataset, and ranked on both public and private leaderboards in real time. The top five participants with the highest ranking on the private leaderboard will be invited to submit AI models and technical reports describing their solutions before the deadline. These solutions will be reviewed and validated by an independent technical review committee consisting of domain experts. The top three winners will be invited to showcase their solutions at the IJCAI conference and awarded cash prizes.

ELIGIBILTY

The competition is open to individuals worldwide who are 18 years of age or older and have reached the age of majority in their country of residence.

Submission Rules

Each participant can only register one account and participate in one team. Each team can have at most 5 members. If a participant does not join a team, he/she shall be considered a team.

The competition’s registration and submission portal will be open on the challenge start date.

Dataset

  • The training dataset consists of 10,000 street-level monocular images (in .jpg format) with corresponding camera pose data (in .csv format) as ground truth.
  • The testing dataset contains 2,000 street-level monocular images (in .jpg format), with an unreleased partitioning for the public and private leaderboards, respectively.

PRIZES

The top three teams will be declared as the winners of the challenge and awarded the following cash prizes:

  • First Place: USD 20,000
  • Second Place: USD 10,000
  • Third Place: USD 5,000

Check out the winners here.

TERMS AND CONDITIONS

The terms and conditions are available here.

JOINTLY ORGANISED BY:

HOSTED BY:

For more questions, please email to prizechallenge@aisingapore.org.