Tobias Fischer
@TobiasFischer11
Research Scientist Intern @NVIDIA. PhD student at CVG, @ETH Zurich. Previously @Meta @RealityLabs, @RWTH, @berkeley_ai
📢 Introducing GenSfM – a generic Structure-from-Motion pipeline built on COLMAP, with smooth, invertible, and universal calibration for diverse cameras — from pinhole to catadioptric! Thanks @don2qian @visionviktor @mapo1 for the great collaboration. 🔗 ivonne320.github.io/yihan.home/Gen…
SplatAD: Real-Time Lidar and Camera Rendering with 3D Gaussian Splatting for Autonomous Driving Georg Hess, Carl Lindström, Maryam Fatemi, Christoffer Petersson, Lennart Svensson tl;dr: 3DGS->rolling shutter+lidar intensity+ray dropping+variations arxiv.org/pdf/2411.16816
Great opportunity to build cool stuff with brilliant minds.
📢📢We have a last-minute internship opening on my team at @NVIDIAAI for this summer. If you are interested and have experience with large feedforward reconstruction models or post-training image/video diffusion models, please get in touch!
Monocular Dynamic Reconstruction is HARD! Single video input –> A LOT of 3D ambiguity –> Significant visual artifacts. BulletGen solves this using Generative Models, at various frozen ‘bullet-times’, fixing visual artifacts + filling-in missing info == much better results!
Introducing 📢💥BulletGen: Improving 4D Reconstruction with Bullet-Time Generation We improve 4D reconstruction from monocular videos by augmenting with bullet-time reconstructions form a generative model. rozumden.github.io/bulletgen/ arxiv.org/abs/2506.18601
🌊🏄
Accepted to ICCV with perfect review scores!!! See you all in Hawaii!!!!
The time for new architectures is over? Not quite! SeNaTra, a native segmentation backbone, is waiting, let's see how it works 🧵arxiv.org/abs/2505.16993
Watch MP-SfM crush VGGT and Mast3r-SfM! Great work on leveraging monocular priors to tackle some of the most challenging problems in SfM (low-overlap, low-parallax, symmetries) 👏
Excited to share the first paper of my PhD: MP-SfM! #CVPR2025 We augment SfM with monocular depth & normals to handle extreme low-overlap scenes. 🚀 Code: github.com/cvg/mpsfm 📄 Paper: arxiv.org/pdf/2504.20040 Exciting collaboration with @pesarlin, J.L. Schönberger & @mapo1
✨We introduce SuperDec, a new method which allows to create compact 3D scene representations via decomposition into superquadric primitives! Webpage: super-dec.github.io ArXiv: arxiv.org/abs/2504.00992 @BoyangSun @FrancisEngelman @mapo1 @cvg_ethz @ETH_AI_Center
As we learn to use compute more efficiently, we will use more compute.
Today OpenAI announced o3, its next-gen reasoning model. We've worked with OpenAI to test it on ARC-AGI, and we believe it represents a significant breakthrough in getting AI to adapt to novel tasks. It scores 75.7% on the semi-private eval in low-compute mode (for $20 per task…
That‘s a wrap! Thanks to the organizers of @NeurIPSConf for putting together an amazing conference. See you soon Vancouver (what a beautiful venue!) 🛫
I’ll present our #NeurIPS2024 Spotlight Poster on Friday, 13 Dec 4:30-7:30pm PST in East Exhibit Hall A-C, booth #1505. Please step by to talk large-scale dynamic 3D Gaussians! :)
Dynamic 3D Gaussian Fields for Urban Areas tobiasfshr.github.io/pub/4dgf/ We optimize a single dynamic scene representation from a set of heterogeneous input sequences of a common geographic area, rendering at interactive speeds while modeling huge appearance and geometry changes.
Introducing 🛹 RollingDepth 🛹 — a universal monocular depth estimator for arbitrarily long videos! Our paper, “Video Depth without Video Models,” delivers exactly that, setting new standards in temporal consistency. Check out more details in the thread 🧵
Happy to share that our paper was accepted as spotlight @NeurIPSConf 2024! See you in Vancouver 🇨🇦
Dynamic 3D Gaussian Fields for Urban Areas tobiasfshr.github.io/pub/4dgf/ We optimize a single dynamic scene representation from a set of heterogeneous input sequences of a common geographic area, rendering at interactive speeds while modeling huge appearance and geometry changes.
Hyperscape: The future of VR and the Metaverse Excited that Zuckerberg @finkd announced what I have been working on at Connect. Hyperscape enables people to create high fidelity replicas of physical spaces, and embody them in VR. Check out the demo app: meta.com/experiences/79…
📣📣 PhD-Intern for Generative AI 📣📣 We are looking for an intern working on 3D Generative AI for summer 2025! Apply here: metacareers.com/jobs/198607184… + feel free to reach out to @K_S_Schwarz or me via PM/mail if you have any questions!
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