The IEEE/CVF Conference on Computer Vision and Pattern Recognition awarded its 2026 Best Paper prize to D4RT, a unified 4D scene-reconstruction model developed by researchers at Google DeepMind, Oxford University and University College London. The award was announced at the CVPR 2026 ceremony in Denver on June 5.
What D4RT actually does
D4RT — short for "efficiently reconstructing dynamic scenes one D4RT at a time" — is a transformer-based architecture that jointly produces depth maps, 3D point tracks and full camera parameters for any dynamic video. Instead of stitching together separate networks for monocular depth, optical-flow tracking and structure-from-motion, the model uses a single encoder-decoder to give independent queries for the 3D position of any point in space and time. The result is a 4D representation that is faster and more memory-efficient than prior baselines on standard CVPR benchmarks.
Why the field is going 4D
D4RT is the latest signal that computer vision is sliding from 2D recognition to 4D world modelling, in lockstep with embodied AI and robotics. CVPR 2026 itself was the largest in the conference's history — 16,092 submitted papers and 4,089 accepted — and generative, multimodal and embodied-AI work nearly doubled their share. The systems that win CVPR awards today increasingly become the perception stack inside next year's autonomous-vehicle and humanoid robot platforms.
Who's on the paper
The paper was led by Chuhan Zhang during her internship at Google DeepMind, with co-authors Guillaume Le Moing, Skanda Koppula, Ignacio Rocco, Liliane Momeni, Junyu Xie, Shuyang Sun, Rahul Sukthankar, Joëlle Barral, Raia Hadsell, Zoubin Ghahramani, Andrew Zisserman, Junlin Zhang and Mehdi S. M. Sajjadi. Andrew Zisserman is a long-time leader of Oxford's Visual Geometry Group and a co-founder of Google's vision research culture; the credit list underscores how tightly DeepMind, Oxford and UCL now work as a single UK-Alphabet vision lab.
For more CVPR 2026 coverage, see our piece on the record 16,092 paper submissions, Apple's 14 papers at the conference, and the opening overview at CVPR 2026 Opens in Denver.
Reporting based on the IEEE Computer Society and CVF awards announcement, the CVPR 2026 Open Access Repository, Newswise and TechTimes coverage.