David Novotny (@davnov134) / X

David Novotny

111 posts

David Novotny

@davnov134

3D GenAI; Ex-Meta AI Research @metaai; previously a DPhil at the VGG Group, University of Oxford @Oxford_VGG

Joined July 2014

Pinned
David Novotny
@davnov134
Apr 28
🚀🌍🚀 Echo-2 is live! It generates high-quality physically grounded 3D scenes instead of just pixels. Check out the open demo to create and explore worlds in real-time! I gave it a picture from my Himalayas trek—it handles the massive scale range really well! 🏔️🌍
00:00
01:01
SpAItial AI
@SpAItial_AI
Apr 28
🚀Echo-2 is here - our new world model! These aren’t videos. These are 𝟑𝐃 𝐬𝐜𝐞𝐧𝐞𝐬. Generated from a single image. - Stunning visual quality. - Real-time rendering. - Interactive camera control. - Physically grounded. 🧵More details👇
11K
David Novotny
@davnov134
May 1, 2024
Excited to release Lightplane: A highly memory-efficient differentiable NeRF renderer and feature splatter. It renders FullHD image batches using <1GB of GPU memory and greatly scales up 3D deep models such as the Large Reconstruction Model. With @AngCao3, @jcjohnss, Andrea
GIF
54K
David Novotny
@davnov134
Feb 24, 2025
We are forging something very new in 3D GenAI for cutting-edge content creation and understanding with 3D Foundational Models! 🔨 ➡️ We're looking for 3D/ML/Infra engineers and scientists in Munich & London. Get in touch for details!
16K
David Novotny
@davnov134
Jan 28, 2025
After more than six amazing years, I have decided to give my badge a well-deserved break and, along with it, I am leaving Meta to start something new. During my time at Meta, I was lucky enough to collaborate on top ML projects in 3D and GenAI across all three major research
17K
David Novotny
@davnov134
Jan 14, 2025
We are releasing uCO3D! Built to supercharge 3D GenAI and digital-twin models, this evolution of CO3D features more and higher-quality object videos from 1k categories, 3D Gaussian Splats, and streamlined OSS tools. 💻Data&code: github.com/facebookresear… 📄Paper:
00:00
17K
David Novotny
@davnov134
Jul 19, 2024
VGGSfM is the new SOTA in structure-from-motion, surpassing Iterative-SfM (COLMAP) after 20 years. This end-to-end trained PyTorch solution outperforms it without using a single line of COLMAP code. Congrats to @jianyuan_wang for this amazing success and winning the IMC
AK
@_akhaliq
Jul 19, 2024
Meta releases VGGSfM Visual Geometry Grounded Deep Structure From Motion Structure-from-motion (SfM) is a long-standing problem in the computer vision community, which aims to reconstruct the camera poses and 3D structure of a scene from a set of unconstrained 2D images.
00:00
18K
David Novotny
@davnov134
Mar 17, 2025
📐 VGGT: The new SOTA in multi-view feedforward 3D reconstruction: - Blazing fast: 100 images reconstructed in <4 sec - SOTA accuracy: Outperforms all existing multi-view feed-forward reconstructors Code & demo & comparisons 👉 vgg-t.github.io
Jianyuan
@jianyuan_wang
Mar 17, 2025
Introducing VGGT (CVPR'25), a feedforward Transformer that directly infers all key 3D attributes from one, a few, or hundreds of images, in seconds! No expensive optimization needed, yet delivers SOTA results for: ✅ Camera Pose Estimation ✅ Multi-view Depth Estimation ✅ Dense
00:00
7.2K
David Novotny
@davnov134
Jul 2, 2024
Excited to release Meta 3D Gen, a marriage of AssetGen with TextureGen. AssetGen is a culmination of our mesh generation effort, starting 2 years ago with Holo(Dif)fusion, and drastically evolving into the state-of-the-art AssetGen relased today. Congratulations to the whole
AI at Meta
@AIatMeta
Jul 2, 2024
📣 New research from GenAI at Meta, introducing Meta 3D Gen: A new system for end-to-end generation of 3D assets from text in <1min. Meta 3D Gen is a new combined AI system that can generate high-quality 3D assets, with both high-resolution textures and material maps end-to-end,
00:00
5.7K
David Novotny
@davnov134
Oct 12, 2021
Really happy to see our CO3D paper receiving the ICCV'21 Honorable Mention Award. Congrats to the whole team!
Kosta Derpanis (sabbatical in Zurich)
@CSProfKGD
Oct 12, 2021
#ICCV2021 paper awards honourable mentions
David Novotny
@davnov134
Sep 2, 2021
Excited to release the Common Objects in 3D (CO3D) dataset together with our @ICCV_2021 paper (arxiv.org/abs/2109.00512). In collaboration with Jeremy Reizenstein, @ovrdr, @philipphenzler (UCL), @ginocornabo, @monsieurlabatut
This post is unavailable.
David Novotny
@davnov134
Jul 19, 2022
Excited to release the CO3D Challenge (eval.ai/web/challenges…) and #CO3Dv2 (github.com/facebookresear…) - a bigger and improved version of the original CO3D.
AI at Meta
@AIatMeta
Jul 19, 2022
The Common Objects in 3D v2 #CO3Dv2 dataset is the next version of the first real-life dataset enabling large-scale 3D reconstruction of object categories. Together w/ the new version, we're also pleased to announce the CO3D challenge happening at #ECCV22! eval.ai/web/challenges…
00:00
David Novotny
@davnov134
Jan 28, 2025
Replying to @davnov134
I am also happy to share that we are now starting a new venture with an incredible founding team. If you're curious about what we’re building next or interested in collaborating, feel free to drop me a message!
1.4K
David Novotny
@davnov134
Jan 14, 2025
Replying to @davnov134
Compared to CO3D, uCO3D contains 20x more categories and 4x more objects. Moreover, each scene is newly annotated with a 3D Gaussian Splat reconstruction, and with a text caption. This enables training large text-to-3D GenAI models, like the one from our paper.
478
David Novotny
@davnov134
Jan 14, 2025
Replying to @davnov134
Moreover, following the community feedback, we: - Increased video quality - 60% of videos are 1080p or higher. - Released raw videos instead of just the frames, including the ones where SfM failed. - Extract & undistort video frames inside the dataloader on-the-fly. - Decreased
729