Skip to content

Meet Déjà View, a model that turns photos taken from different angles into 3D geometry and camera positions. By looping a single transformer block, this 117M-parameter model matches or outperforms models up to 10x larger across five benchmarks, using comparable or less compute. Researchers can dial the number of loops to fit their compute budget. #NeurIPS2026 oral paper + code: http://research.nvidia.com/labs/dvl/projects/dvlt

NVIDIA AI · X· @NVIDIAAI·· 5 days ago

3.5K views · 55 likes · 11 reposts (at collection)

Original
Who
NVIDIA researchers
What
introduce Déjà View, a 117M-parameter model that reconstructs 3D geometry and camera positions from photos taken from different angles
Quote
Meet Déjà View, a model that turns photos taken from different angles into 3D geometry and camera positions.

Read next

See the full storyNVIDIA Researchers Introduce Déjà View for 3D Reconstruction from Photos
Article

Meet Déjà View, a model that turns photos taken from different angles into 3D geometry and camera positions.

By looping a single transformer block, this 117M-parameter model matches or outperforms models up to 10x larger across five benchmarks, using comparable or less compute. Researchers can dial the number of loops to fit their compute budget.

#NeurIPS2026 oral paper + code: research.nvidia.com/labs/dvl/…

The story so far

NVIDIA researchers introduced Déjà View, a 117-million-parameter model that reconstructs 3D geometry and camera positions from photos taken at different angles. The researchers said it matched or outperformed models up to ten times larger on five benchmarks while using comparable or less compute. Déjà View loops a single transformer block, and researchers can adjust the number of loops to fit their compute budget. NVIDIA presented the work as a NeurIPS 2026 oral paper and shared its code.