Hanwen Jiang
@hanwenjiang1
Research Scientist @ Adobe Research | PhD @ UT Austin | I work on 3D Vision
Supervised learning has held 3D Vision back for too long. Meet RayZer — a self-supervised 3D model trained with zero 3D labels: ❌ No supervision of camera & geometry ✅ Just RGB images And the wild part? RayZer outperforms supervised methods (as 3D labels from COLMAP is noisy)…
Hope you enjoyed our workshop 😛 We have now released the slides of the presentations. Thanks to the amazing speakers! Check it if you didn't attend!
🔍 3D is not just pixels—we care about geometry, physics, topology, and functions. But how to balance these inductive biases with scalable learning? 👀 Join us at Ind3D workshop @CVPR (June 12, afternoon) for discussions on the future of 3D models! 🌐 ind3dworkshop.github.io/cvpr2025
Talk by @KostasPenn happening now!
🔍 3D is not just pixels—we care about geometry, physics, topology, and functions. But how to balance these inductive biases with scalable learning? 👀 Join us at Ind3D workshop @CVPR (June 12, afternoon) for discussions on the future of 3D models! 🌐 ind3dworkshop.github.io/cvpr2025
Top minds. Deep ideas. 🎯 Inductive Bias in 3D Generation 🗓️ June 12 — Day 2 of CVPR. Be there!
🔍 3D is not just pixels—we care about geometry, physics, topology, and functions. But how to balance these inductive biases with scalable learning? 👀 Join us at Ind3D workshop @CVPR (June 12, afternoon) for discussions on the future of 3D models! 🌐 ind3dworkshop.github.io/cvpr2025
Our paper LVSM has been accepted as an oral presentation at #ICLR2025! See you in Singapore! We’ve just released the code and checkpoints—check it out here: github.com/haian-jin/LVSM.🚀
Novel view synthesis has long been a core challenge in 3D vision. But how much 3D inductive bias is truly needed? —Surprisingly, very little! Introducing "LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias"—a fully transformer-based approach that enables scalable,…
Working on Depth Estimation? Here is a free lunch. We tune a Depth Anything ViT-B model on MegaSynth, and the performance improves a lot -- depth estimation is also very non-semantic! #CVPR2025 Accepted
💥 Think more real data is needed for scene reconstruction? Think again! Meet MegaSynth: scaling up feed-forward 3D scene reconstruction with synthesized scenes. In 3 days, it generates 700K scenes for training—70x larger than real data! ✨ The secret? Reconstruction is mostly…
Want to generate INFINITE 3D Gaussians? Check Atlas Gaussian -- we decode Gaussians as local geometry patches. With sampling in the local UV space, you can decode infinite Gaussians! Congrats to Haitao (yanghtr.github.io) for AtlasGaussian being selected as spotlight at…
Very happy that AtlasGaussians was accepted by ICLR 25. openreview.net/forum?id=H2Gxi…). I did very little and the students, in particular the first author Haitao Yang (yanghtr.github.io), came up with the idea. Haitao is graduating soon. Also, my second published paper with…