Nvidia's Tiny 3D AI Matches Models 10 Times Its Size By Reusing One Layer

Nvidia researchers, working with ETH Zürich, the University of Toronto and the University of Modena and Reggio Emilia, built Déjà View, a 117-million-parameter model that reconstructs 3D scenes and camera poses from multi-angle photos by running the same transformer block repeatedly instead of stacking distinct layers. Across five benchmarks covering indoor rooms, outdoor scenes, single objects and driving footage, the smallest model tested led on average camera pose accuracy, matching or beating models up to 10 times larger such as Depth Anything 3 (1.2B parameters) and VGGT (1.26B). The looped design also outperformed an otherwise identical version with separate weights per step trained on the same data and compute, and users can choose the number of passes to trade speed for accuracy.

Nvidia researchers have built Déjà View, a 117 million-parameter AI model that turns photos taken from different angles into 3D scenes and matches or beats models up to 10 times its size by reusing a single layer over and over. Nvidia's AI research team shared https://x.com/NVIDIAAI/status/2120617173286883692 the model on Wednesday. The work comes from 12 researchers at Nvidia, ETH Zürich, the University of Toronto and the University of Modena and Reggio Emilia. Most 3D reconstruction models have grown past 1 billion parameters by stacking many different layers. Déjà View instead runs the same transformer block again and again on features from each photo, sharpening its estimate of the scene and camera positions with every pass. Users can choose how many passes to run from one trained model, trading speed for accuracy depending on how much computing power they have. Nvidia says the looped design even beat an otherwise identical version with separate weights for each step, trained on the same data with the same compute. Also Read: Humanoid Robots Running Google’s Gemini Just Did Factory Work At Mercedes-Benz, And Got 96% Right https://www.fathom.news/gemini-robotics-mercedes/ On five benchmarks covering indoor rooms, outdoor scenes, single objects and driving footage, Déjà View led on average camera pose accuracy while being the smallest model tested. Its rivals included Depth Anything 3 at 1.2 billion parameters, VGGT at 1.26 billion and π³ at 959 million. Read Next: Meta’s AI Code Checker Was Barely Better Than Chance, Then Its Engineers Tuned It https://www.fathom.news/ai-code-judge/ Originally published on Fathom News https://www.fathom.news/nvidia-3d-ai-model/ .