NVIDIA Déjà View turns multi-angle photos into 3D geometry
Weight-tied looping shows parameter efficiency, not scale, can dominate in 3D vision tasks, which shifts how small labs should think about model budgets. Teams doing photogrammetry, robotics perception or 3D content generation can deploy it on modest hardware and tune loop count to their compute ceiling. Code is available via NVIDIA's research page, making it directly testable against larger pose-estimation baselines.
NVIDIA introduced Déjà View, a 117M-parameter model that reconstructs 3D geometry and camera positions from photos taken at different angles. It was presented as an oral paper at NeurIPS 2026, with code released on NVIDIA's research site.
The key design choice is architectural: instead of stacking many transformer layers, Déjà View loops a single transformer block repeatedly. By adjusting the number of loops, researchers can trade accuracy against compute at inference time, and the model matches or beats models up to 10 times larger on five benchmarks while using comparable or less compute.
The result challenges the assumption that parameter count drives reconstruction quality in pose estimation and 3D reconstruction. For practitioners with limited GPU budgets, weight-reuse architectures like this offer a path to strong spatial AI performance on smaller hardware footprints.
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/proje…
This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include NVIDIA, Déjà View, 3D reconstruction, NeurIPS 2026.
Source attribution: curated from NVIDIA AI (@NVIDIAAI). Cross-source corroboration: 11 independent sources have covered this development (including rohanpaul_ai, Together AI, IT之家), confirming industry-wide signal strength.
