From the source
NVIDIA introduced Déjà View, a 117M-parameter model that turns photos taken from different angles into 3D geometry and camera positions.
- The model loops a single transformer block, and NVIDIA says it matches or outperforms models up to 10x larger across five benchmarks while using comparable or less compute.
- Researchers can dial the number of loops to fit their compute budget, according to the company.
- The work is an oral paper at NeurIPS 2026, and NVIDIA says code is available alongside it.






