AI Judge changed title after evaluation, original title: "Chuhan Zhang wins CVPR 2026 Best Paper for D4RT, a unified 4D reconstruction model that speeds up pose estimation 100x"
Chuhan Zhang led work on D4RT while interning at Google DeepMind, earning the top CVPR 2026 prize for a single transformer that reconstructs dynamic scenes from ordinary video. The model jointly produces depth maps, 3D tracks, and camera poses through one encoder-decoder instead of separate networks for each job.
Kosta Derpanis (sabbatical in Zurich)#904
The award ceremony is going to start at the #CVPR2026 conference!
Who will take home the honors? 🏆
Check out the 74 paper award candidates!
https://cvpr.thecvf.com/virtual/2026/events/AwardCandidates2026
Rather than decoding dense outputs every frame, D4RT lets users probe any 3D location in space-time on demand. This design collapses tracking, depth, and pose into the same forward pass and removes the need for multiple specialized heads.
Reported throughput reaches roughly 200 frames per second for pose on relevant benchmarks and processes a minute of video in seconds on one TPU. No public code or weights have been confirmed yet, so the gap between benchmark gains and everyday robotics or AR pipelines stays untested.
Many users congratulated the D4RT team on winning the CVPR 2026 Best Paper for unified 4D scene reconstruction, while some criticized DeepMind-linked work for lacking reproducibility and code.
Announcing the official #CVPR2026 Best Paper Award Winner! 🏆 Congratulations to the authors for their landmark contributions to the field!👏👏
Congrats to the team for wining CVPR Best Paper Award!! 🏆 Come to our oral session (Mile High Ballroom 13:00-14:15) and poster (16:00-18:00) today for more details 🚀
A SINGLE encoder + decoder for all the 4D tasks! We release 🎯 D4RT (Dynamic 4D Reconstruction and Tracking).
📍 A simple, unified interface for 3D tracking, depth, and pose 🌟 SOTA results on 4D reconstruction & tracking 🚀 Up to 100x faster pose estimation than prior works
Huge congrats to the team, D4RT is a team work and all the authors have been working very hard on this in the past one year. Very well deserved. 🍻 and thank you Award Committee Members for the recognition.
Announcing the official #CVPR2026 Best Paper Award Winner! 🏆 Congratulations to the authors for their landmark contributions to the field!👏👏
We finally caught up with the star of D4RT @ChuhanZhang5 ❤️ Visit the very packed poster number 20 @CVPR
Huge huge congrats @ChuhanZhang5 @skandakoppula @GuillaumeMoing Ignacio @LiliMomeni @JunyuXieArthur Shuyang, Rahul @joelle_barral @RaiaHadsell @ZoubinGhahrama1 AZ, Junlin &Mehdi @GoogleDeepMind for winning Best Paper @CVPR #CVPR2026... Thrilled for you all 🥳
Huge huge congrats @ChuhanZhang5 @skandakoppula @GuillaumeMoing Ignacio @LiliMomeni @JunyuXieArthur Shuyang, Rahul @joelle_barral @RaiaHadsell @ZoubinGhahrama1 AZ, Junlin &Mehdi @GoogleDeepMind for winning Best Paper @CVPR #CVPR2026... Thrilled for you all 🥳
The PAMI TC awards are here #CVPR2026. Congratulations to all the winners for your contribution to the research community 🏆
Had no idea the CVPR test of time award was named after Longuet-Higgins. He was a chemical physicist who pivoted to cognitive science and was Geoff Hinton's PhD advisor. I came across his work on vibronic coupling when working on quantum excited states. Quite a career.
ResNet and YOLO received to the Longuet-Higgins Test of Time award. Congrats! Three thoughts: - very “difficult” job for the committee this year - people are still using both quite a bit - time for an additional generation to feel old — I already got used to that 😅 #cvpr2026
@pesarlin 100%, reproducibility should be a requirement. Unofficial open-source implementation: https://github.com/Lijiaxin0111/Open-d4rt
@pesarlin They have a really fancy project page though, what more could you want?
A delightful surprise! Great work by the team pushing the frontier of unified 4D perception: estimating depth, camera pose, 3D point tracks, and point clouds from videos.
Announcing the official #CVPR2026 Best Paper Award Winner! 🏆 Congratulations to the authors for their landmark contributions to the field!👏👏
Honorable Mentions👏
Congratulations to @ChuhanZhang5 @LiliMomeni @JunyuXieArthur Andrew and the team!! 🥳🥳
Huge congrats to the team, D4RT is a team work and all the authors have been working very hard on this in the past one year. Very well deserved. 🍻 and thank you Award Committee Members for the recognition.
@pesarlin Technically, I like the paper, the idea and results are super promising. I also tried to reproduce it (without using a coding agent), but the paper really lacks of enough details to reproduce😅
Finalists 👏
@pesarlin When >30k papers are submitted at any AI conf, having a code should be mandatory
@pesarlin Classic DeepMind pattern: elegant work, zero reproducibility. Talent is real but the ecosystem stays closed. That's not how science advances.
If I remember correctly, it was a double-blind review process, so authors should not have added their repo or anything else that would allow reviewers to identify the paper as theirs. I am sure this was done to eliminate favoritism. Once your paper is accepted, you can disclose anything—the code and so on.
@WenCabrel_ @pesarlin there are ways to anonymize code repositories before publicly release it after acceptance, but that's not the issue here. The paper is now accepted with a project page and blog post, but still no code.
@pesarlin how can we know if anything in this paper real, if we are unable to reprod? difference between science and product gets blurrier every day
@yu47257737 @pesarlin You should always read my tweets in an annoyingly sarcastic voice
