David Rozenberszki

📢📢📢We've released the ScanNet++ Novel View Synthesis Benchmark for iPhone data! 🥳 Test your models on RGBD video featuring real-world challenges like exposure changes & motion blur! Download the newest iPhone NVS test split and submit your results! ⬇️ scannetpp.mlsg.cit.tum.de/scannetpp/benc…

📢 SceneComp @ ICCV 2025 🏝️ 🌎 Generative Scene Completion for Immersive Worlds 🛠️ Reconstruct what you know AND 🪄 Generate what you don’t! 🙌 Meet our speakers @angelaqdai, @holynski_, @jampani_varun, @ZGojcic @taiyasaki, Peter Kontschieder scenecomp.github.io #ICCV2025

🎉Meet the #3DV2026 Keynote Speakers! Jitendra Malik · University of California, Berkeley Angela Dai · Technical University of Munich Christian Rupprecht · University of Oxford Alec Jacobson · University of Toronto Bring your latest work and join us for the exciting keynotes!

Check out our #ICCV2025 work on functional 3d scan editing, learning to optimize, multi-level 3d captioning, interactive mesh editing, audio-driven avatars, & shape matching! Congrats @ElBoudjogh24002, @liuyuehcheng, @chandan__yes, @hcxrli, @shivangi2201, Emery for amazing work!

📢 IntrinsiX: High-Quality PBR Generation using Image Priors 📢 From text input, we generate renderable PBR maps! Next to editable image generation, our predictions can be distilled into room-scale scenes using SDS for large-scale PBR texture generation. We first train separate LoRA modules for the intrinsic properties of albedo, rough/metal, normal. Then, we introduce cross-intrinsic attention using a rerendering loss with importance-weighted light sampling to enable coherent PBR generation. Our method outperforms text -> image -> PBR methods both in generalization and quality, since directly generating PBR maps does not suffer from the inherent ambiguity of intrinsic image decomposition. In addition, our design choice facilitates SDS-based PBR texture distillation. 🌍 peter-kocsis.github.io/IntrinsiX/ 🎥 youtu.be/b0wVA44R93Y Great work by @Peter4AI, @LukasHollein

📢ExCap3D: Multilevel Captioning of Objects in 3D Scenes @chandan__yes generates consistent object and part-level descriptions of objects in 3D scenes, and introduces a new dataset with 190k captions for 34k ScanNet++ objects. Project: cy94.github.io/excap3d w/ @david_roz_

Check out our #CVPR2025 papers on articulated mesh generation, 4d shape generation with dictionary neural fields, large-scale 3d scene generation and editing, and 3d editing! Congrats to @DaoyiGao, @xinyi092298, @ABokhovkin, @QTDSMQ, @ErkocZiya for their amazing work!

Excited to bring back the 2nd ScanNet++ Novel View Synthesis and 3D Semantic Understanding Challenge @CVPR 2025! Submit to our new benchmark, now with over 1000 high-fidelity scans of real-world scenes: kaldir.vc.in.tum.de/scannetpp/ Check it out: kaldir.vc.in.tum.de/scannetpp/cvpr…

Excited to announce ScanNet++ v2!🎉 @chandan__yes and @liuyuehcheng have been working tirelessly to bring: 🔹1006 high-fidelity 3D scans 🔹+ DSLR & iPhone captures 🔹+ rich semantics Elevating 3D scene understanding to the next level!🚀 w/ @MattNiessner kaldir.vc.in.tum.de/scannetpp

📢MeshArt: Generating Articulated Meshes with Structure-guided Transformers @DaoyiGao generates articulated meshes with a hierarchical transformer, modeling articulation-aware structures that guide mesh synthesis. w/ @yawarnihal @craigleili Project: daoyig.github.io/Mesh_Art/

Super happy to present our #NeurIPS paper 𝐂𝐨𝐡𝐞𝐫𝐞𝐧𝐭 𝟑𝐃 𝐒𝐜𝐞𝐧𝐞 𝐃𝐢𝐟𝐟𝐮𝐬𝐢𝐨𝐧 𝐅𝐫𝐨𝐦 𝐚 𝐒𝐢𝐧𝐠𝐥𝐞 𝐑𝐆𝐁 𝐈𝐦𝐚𝐠𝐞 in Vancouver. Come to our poster #2804 on Wednesday 11am - 2pm in East Exhibit Hall A-C and say hi if you want to learn more about 3D Scene Diffusion! See you tomorrow! Project Page: manuel-dahnert.com/research/scene… Poster: neurips.cc/virtual/2024/p…

Congrats to the team @pablorpalafox 🥳

📢SceneFactor: Generating & editing 3D indoor scenes from text! @ABokhovkin presents a factored latent diffusion for controllable, large-scale scene synthesis -- decomposed into high-level semantic generation + geometric refinement w/ @QTDSMQ, @shubhtuls alexeybokhovkin.github.io/scenefactor/

Congrats to @Normanisation who won the best paper award of @MdsiTum for his DiffRF work! DiffRF is among the first 3D generative models that leverages diffusion, and that directly operates on a 3D radiance field representation.

(1/2) How to accelerate the reconstruction of 3D Gaussian Splatting? 3DGS-LM replaces the commonly used ADAM optimizer with a tailored Levenberg-Marquardt (LM). => We are 𝟑𝟎% 𝐟𝐚𝐬𝐭𝐞𝐫 𝐭𝐡𝐚𝐧 𝟑𝐃𝐆𝐒 for the same quality. lukashoel.github.io/3DGS-LM/ youtu.be/tDiGuGMssg8

Life is probabilistic, and so should be understanding 3D scenes! Check out this really cool work by @DaoyiGao

Check out MultiDiff #CVPR2024! From a single RGB image, MultiDiff enables scene-level novel view synthesis with free camera control. sirwyver.github.io/MultiDiff youtu.be/SKpFFtVJo9c Great work by @normanisation @K_S_Schwarz @barbara_roessle, L Porzi, S Rota Bulò, P Kontschieder

@MattNiessner @Kaplanyan No doubt this was the best part of the entire week!

(1/2) The definite highlight at #CVPR was our 𝐛𝐨𝐚𝐭 𝐭𝐫𝐢𝐩 organized by @Kaplanyan With perfect weather, we started on Mercer Island, boated from Lake Washington to Lake Union, and made our way back by the shores of Medina before heading out to a legendary afterparty.

Come and chat with us at our #CVPR2024 poster UnScene3D to learn how to understand 3D scenes without ANY human annotation. See you tomorrow 10:30 at #40 in Arch. Work done with @angelaqdai and @orlitany. Check out more about our paper at rozdavid.github.io/unscene3d

Excited to present GenZI at #CVPR2024! @craigleili introduces GenZI, the first zero-shot approach to creating realistic 3D human-scene interactions by leveraging interaction priors from large VLMs. Code and data on our website! craigleili.github.io/projects/genzi/ youtu.be/ozfs6E0JIMY