ROOMELSA: Retrieval of Optimal Objects for
Multi-modal Enhanced Language and Spatial Assistance

🗞️ News

📢 We release the Public Dataset comprising 1,622 scenes, 5,197 rooms, and 44,445 object-query pairs.

📢 Team Registration is Now Closed!

📦 The private dataset and submission guidelines have been sent to all registered participants.

🏆 Out of 18 incredible participants, the top 5 have been selected based on performance. Check out the leaderboard to see who made it to the top. Congratulations to all!

ROOMELSA Grand Challenge

The ROOMELSA Grand Challenge presents a novel and impactful task set in 3D environments: retrieving a masked object in a 3D scene using only a natural language description and spatial context.

By framing this retrieval task within a 3D spatial setting, ROOMELSA advances research at the intersection of scene understanding, cross-modal retrieval, and spatial-language alignment. It challenges participants to build intelligent, context-aware systems capable of reasoning across modalities in complex, real-world-like environments.

Leaderboards

We’re excited to share the final leaderboard for this challenge! Out of 18 talented participants, the top 5 have emerged based on performance metrics. Congratulations to all who took part - your efforts made this a truly competitive and inspiring event!