3D Scene Understanding for Vision, Graphics, and Robotics
CVPR 2023 Workshop, Vancouver, June 18th, 2023
Challenge
This year we establish a new challenge on embodied scene understanding featuring the recently introduced SQA3D benchmark. There are two tasks in the challenge:- Situation understanding: Given a scene and a sentence that describes the situation of an agent in that scene, locate the position and orientation of the agent.
- Situated reasoning: Given a scene and a sentence that describe the situation of an agent in that scene, answer a question. You may use the ground truth location in this task.
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.
Additional information
- Since the test set is publicly available, please submit your result directly to the online leaderboard hosted by paperwithcode. More information: SQA3D website
- 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.
- 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.
- If you have more questions, please contact Xiaojian Ma.