3D Scene Understanding for Vision, Graphics, and Robotics

CVPR 2023 Workshop, Vancouver, June 18th, 2023


This year we establish a new challenge on embodied scene understanding featuring the recently introduced SQA3D benchmark. There are two tasks in the challenge:

For more information on SQA3D, please check out the overview slides and the technical report.

The deadline of submitting your result is June 10 2023. The winner will be announced on June 11 2023.

The participants are provided with training, validation and testing sets and an automatic evaluation script. A codebase with baseline models is also available. The winner of each task will be invited to give a short talk describing their method during the workshop.

Submission guidelines

  1. Since the test set is publicly available, please evaluate your model directly on it and send the result to xiaojian.ma@ucla.edu.
  2. There are three different scene representations offered by SQA3D: 3D scan, egocentric video and bird-eye view picture. Please include the representation you use when submitting your result.
  3. Using the ground truth location annotations is allowed in the situated reasoning task, but please add a note of using them when submitting your result.
  4. The submitted result will be made public immediately to the online leaderboard hosted by paperwithcode.
  5. If you have more questions, please contact Xiaojian Ma.