Biao Zhang

🚀 Do world models need explicit 3D? Our answer: if you’re using Transformers, introducing 3D into DiT’s positional encoding is a natural choice. 📄 Paper: arxiv.org/pdf/2510.20385 🌐 HomePage: yunpeng1998.github.io/PE-Field-HomeP… 💻 Code: github.com/MTLab/PE-Field

📢V2M4 (Accepted by ICCV 2025) We reconstruct 4D mesh animations from a single video with consistency in pose, appearance, topology, and texture. project: windvchen.github.io/V2M4/ arxiv: arxiv.org/abs/2503.09631 Collaboration with @_BiaoZhang , Xiangjun Tang and @peter_wonka

@maxxxzdn Great work! Nice to see there are still new networks for point cloud data! I have a question. Is it similar to this work if we follow PointNet++'s grouping strategy and apply local attention in a small neighborhood?

🤹 New blog post! I write about our recent work on using hierarchical trees to enable sparse attention over irregular data (point clouds, meshes) - Erwin Transformer. blog: maxxxzdn.github.io/blog/erwin/ paper: arxiv.org/abs/2502.17019 Compressed version in the thread below:

One of the biggest bottlenecks in deploying visual AI and computer vision is annotation, which can be both costly and time-consuming. Today, we’re introducing Verified Auto Labeling, a new approach to AI-assisted annotation that achieves up to 95% of human-level performance while cutting labeling costs by up to 100,000x and time by 5,000x. Read the full paper: arxiv.org/abs/2506.02359

🔍 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

@isskoro It looks like scale augmentation?

In the past 1.5 weeks, there appeared 2 papers by 2 different research groups which develop the exactly same (and embarrassingly simple) trick to improve convergence of image/video diffusion models by 20-100+% (sic!) arxiv.org/abs/2502.14831 arxiv.org/abs/2502.09509

📢𝐆𝐞𝐨𝐦𝐞𝐭𝐫𝐲 𝐃𝐢𝐬𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐨𝐧𝐬 We propose to represent general geometries as distributions using diffusion models. project site: 1zb.github.io/GeomDist/ arxiv: arxiv.org/abs/2411.16076 dataset: huggingface.co/datasets/Zbalp… Collaboration with Jing Ren and @peter_wonka

@MattNiessner Congratulations!

Official promotion to 𝐅𝐮𝐥𝐥 𝐏𝐫𝐨𝐟𝐞𝐬𝐬𝐨𝐫 🥳🥳🥳 Thanks so much to everyone who has been with me on this exciting journey, in particular my amazing PhD students, PostDocs, collaborators, every other supporter! Let's see what's next - just getting started :)

Happy to share that our Vivid-Zoo paper has been accepted at #NeurIPS2024. The paper is collaborated with @1zhengcheng @WenxuanZhu75798 @_BiaoZhang @peter_wonka @BernardSGhanem

Our #CVPR2024 workshop on 3D foundation models (@ summit 434) is happening now! The workshop is super popular, with a full room of audience and a long line of people eager to enter!

Join us at our first workshop on 3D Foundation Models @CVPR2024, June 18 in Summit 434, starting at 8:50AM! We have fantastic speakers to discuss the progress and prospects in 3D foundation models. Check out more details at 3dfm.github.io

@davech2y 太高产了！

Proudly present SceneTex! If you can describe the style, you can texture the 𝐰𝐡𝐨𝐥𝐞 𝐬𝐜𝐞𝐧𝐞. Check out our paper, code, and results -> daveredrum.github.io/SceneTex

.@_BiaoZhang presenting our 3DShape2VecSet paper at #SIGGRAPH2023!!! We propose a 3D representation for neural fields and generative diffusion models for figh-fidelity shape generation. Super cool work from @peter_wonka's group in collab with @jiapeng_tang.

I will give a talk at 13:45 at the ScanNet workshop today #CVPR2023 to talk about data structures for 3d generative shape modeling. Will discuss some recent work in the field including our own work led by @_BiaoZhang in collaboration with TUM (@MattNiessner @jiapeng_tang )

(1/2) Check out the overview of our #SIGGRAPH2023 works! niessnerlab.org/publications.h… We have 4 accepted papers (3 journal) in exciting areas: we work on dynamic NeRFs for human bodies and faces, a well as 3D diffusion models for shape generation and text-based content generation!

Also Thanks @_akhaliq for tweet twitter.com/_akhaliq/statu…

Today we released our #CVPR2023 paper in collaboration with Meta: VIVE3D: Viewpoint-Independent Video Editing using 3D-Aware GANs Anna Frühstück (@farnunea), Nikolaos Sarafianos (@sarafianosn), Yuanlu Xu, Peter Wonka, Tony Tung arxiv.org/abs/2303.15893 afruehstueck.github.io/vive3D/

#CVPR2023 paper is online (updated) now: 3DAvatarGAN: Bridging Domains for Personalized Editable Avatars @AbdalRameen, @hyjameslee, Peihao Zhu, Menglei Chai, Aliaksandr Siarohin, @peter_wonka, @SergeyTulyakov arxiv.org/pdf/2301.02700… rameenabdal.github.io/3DAvatarGAN/

Check out our papers at #ECCV in Tel Aviv - we'd love to chat and talk about the latest research :) Great work by @yunluchen111, @yawarnihal, @DiChang10, @YujinChen_cv , @shivangi2201, @yinyu_nie and collaborators! Poster sessions below in🧵

Happy to share that our NSDP got accepted to #NeurIPS2022 ! Homepage: tangjiapeng.github.io/projects/NSDP/ We proposed transformer-based deformation networks to learn realistic non-rigid deformation priors. Many thanks to my collaborators : Lev, Bi, @JustusThies @MattNiessner