Chuanxia Zheng
@ChuanxiaZ
Incoming Nanyang Assistant Professor @NTUsg | MSCA Fellow @MSCActions | Postdoc @Oxford_VGG | PhD @NTUsg
After two amazing years with @Oxford_VGG, I will be joining @NTUsg as a Nanyang Assistant Professor in Fall 2025! I’ll be leading the Physical Vision Group (physicalvision.github.io) — and we're hiring for next year!🚀 If you're passionate about vision or AI, get in touch!
Thanks @_akhaliq for twittering our work. Our CoPart (accepted at ICCV2025 @ICCVConference) proposes a part-based 3D generation framework & the first large-scale manually-annotated part-level 3D dataset named PartVerse. It contains 12k objects and 91k parts, with part priority…
From One to More Contextual Part Latents for 3D Generation
🚨 PhD opportunity! I’m recruiting a PhD student with a strong background in math/stats/CS to join my group at NTU Singapore 🇸🇬 Start: Jan or Aug 2026 Topic: Foundations of Epistemic Uncertainty in ML 🧠🔍 📌 Details: chau999.github.io/group/ RT appreciated!
【Call for Reviewers】Please consider to join the #ICCV2025 LIMIT Workshop as a reviewer! iccv2025-limit-workshop.limitlab.xyz - Review assignment: 1-2 papers - Review Period: 19 Jul - 6 Aug, 2025 - Form: forms.gle/e7Pvu8w3L57Eke… <= apply here! We appreciate your support in LIMIT 2025!
Amodal3R is accepted at @ICCVConference !!! A method for completed 3D reconstruction from only partially visible inputs. Thanks to the amazing team.
Amodal3R: Amodal 3D Reconstruction from Occluded 2D Images TL;DR: Given partially visible objects within images, Amodal3R reconstructs semantically meaningful 3D assets with reasonable geometry and plausible appearance.
Geo4D is accepted at @ICCVConference! A method to repurpose video diffusion models for monocular 4D reconstruction. Thanks to the amazing team: @CodyJzr, Iro Laina, @dlarlus, and Andrea Vedaldi @Oxford_VGG
🎁 We present Geo4D, a method that repurposes a video diffusion model for monocular 4D reconstruction. Project page: geo4d.github.io Code repo: github.com/jzr99/Geo4D 𝐌𝐚𝐢𝐧 𝐂𝐨𝐧𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐨𝐧𝐬: ✨ A novel framework, Geo4D, to reconstruct the dynamic scene,…
DSO is accepted at @ICCVConference! A feed-forward 3D generator that directly outputs physically stable objects, even for in the real world Huge thanks to the amazing team @Oxford_VGG: @RayLi234, Christian Rupprecht, and Andrea Vedaldi!
DSO is out on Hugging Face Aligning 3D Generators with Simulation Feedback for Physical Soundness
Puppet-Master is accepted at @ICCVConference! A physically interactive video generative model for part-level motion. Huge thanks to the amazing team @Oxford_VGG: @RayLi234, Christian Rupprecht, and Andrea Vedaldi!
Puppet-Master Scaling Interactive Video Generation as a Motion Prior for Part-Level Dynamics discuss: huggingface.co/papers/2408.04… We present Puppet-Master, an interactive video generative model that can serve as a motion prior for part-level dynamics. At test time, given a single…
Many Congratulations to @jianyuan_wang, @MinghaoChen23, @n_karaev, Andrea Vedaldi, Christian Rupprecht and @davnov134 for winning the Best Paper Award @CVPR for "VGGT: Visual Geometry Grounded Transformer" 🥇🎉 🙌🙌 #CVPR2025!!!!!!
Join us for the 4D Vision Workshop @CVPR on June 11 starting at 9:20am! We'll have an incredible lineup of speakers discussing the frontier of 3D computer vision techniques for dynamic world modeling across spatial AI, robotics, astrophysics, and more. 4dvisionworkshop.github.io
🤖 Do you have a PhD, and want to push the frontier of computer vision and robotics? 🤖 The Visual Geometry Group (VGG) in Oxford is hiring a postdoc! PI: Dr. João Henriques. Deadline: 2 June at noon (UK). More details: eng.ox.ac.uk/jobs/job-detai…
Play 4D scenes part 2. With the same monocular video input, Geo4D (github.com/jzr99/Geo4D) can now provide a more robust and clear 4D reconstruction result. CRAZYYY. I cannot imagine what is next. 4DGS from monocular video? I think it's feasible already.
We've released Geo4D, a diffusion model that directly rebuilds the dynamic 3D point maps from a monocular video, achieving much more consistent video depth than existing SOTA results. This work is done with @CodyJzr, Iro Laina, @dlarlus, and Andrea Vedaldi @Oxford_VGG
🎁 We present Geo4D, a method that repurposes a video diffusion model for monocular 4D reconstruction. Project page: geo4d.github.io Code repo: github.com/jzr99/Geo4D 𝐌𝐚𝐢𝐧 𝐂𝐨𝐧𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐨𝐧𝐬: ✨ A novel framework, Geo4D, to reconstruct the dynamic scene,…
Thanks @_akhaliq for sharing!!! The task and results are meaningful for 3D design. More importantly, designing a differentiable loss for many physical attributes is hard. The proposed DSO can be a good chance for many physical generation tasks.
DSO is out on Hugging Face Aligning 3D Generators with Simulation Feedback for Physical Soundness
Existing 3D models improve geometry & appearance, but often neglect their physical soundness. 📢📢📢 Introducing DSO, a feed-forward 3D generator that directly outputs physically stable objects, and even works in real life when 3D printed. project page: ruiningli.com/dso
Introducing DSO, a method for aligning 3D generators with simulation feedback for physical soundness. At inference time, DSO-finetuned models are more likely to generate physically sound, self-supporting objects in a feed-forward manner. ⚡️Ultra-fast: no test-time optimization…
Yes‼️Introducing one of the new features of #Rodin Gen-2: Generate into Parts🔥 🤯For the first time, 3D GenAI truly understands the underlying structure of objects. Coming June 2025🚀 Demo made with @huggingface @gradio #3D #MCP #b3d #Game #UE #Unity #GenerativeAI
🤔Quick question: Do Large Models Dream of 3D Relationships? #RodinGen2 #CG #3D
Guess you probably already knew that we have an update for V3!
🚀 DeepSeek-V3-0324 is out now! 🔹 Major boost in reasoning performance 🔹 Stronger front-end development skills 🔹 Smarter tool-use capabilities ✅ For non-complex reasoning tasks, we recommend using V3 — just turn off “DeepThink” 🔌 API usage remains unchanged 📜 Models are…
Hi there, 🎉 We are thrilled to introduce Stable Virtual Camera, a generalist diffusion model designed to address the exciting challenge of Novel View Synthesis (NVS). With just one or a few images, it allows you to create a smooth trajectory video from any viewpoint you desire.…