🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Check out Sherwins great work on adding camera control to video models! 🎥

📣📣 Hiring a PhD-Intern 📣📣 Work with me on Dynamic 3D Gaussians at the Meta Boston office for 6 months in summer 2025! Apply here: metacareers.com/jobs/105497412… + write me your questions / link your most relevant work via email or twitter.

𝗛𝘆𝗽𝗲𝗿𝘀𝗰𝗮𝗽𝗲: 𝗣𝗵𝗼𝘁𝗼𝗿𝗲𝗮𝗹𝗶𝘀𝘁𝗶𝗰 𝗥𝗲𝗽𝗹𝗶𝗰𝗮𝘀 𝗶𝗻 𝗩𝗥 Mark Zuckerberg just unveiled our team's latest project: t.ly/QK7kN ✨Explore digital replicas in VR - captured with a phone 👉 Try it on your Quest (US only): t.ly/BrCrL

A major step towards generalizable 3D reconstruction! Is this another proof that by leveraging tricks from large-context LLMs, we can move away from pure optimization once and for all?

📢𝐋𝟑𝐃𝐆: 𝐋𝐚𝐭𝐞𝐧𝐭 𝟑𝐃 𝐆𝐚𝐮𝐬𝐬𝐢𝐚𝐧 𝐃𝐢𝐟𝐟𝐮𝐬𝐢𝐨𝐧📢 #SIGGRAPHAsia We propose a generative diffusion model for 3D Gaussians. Key is a learnt latent space which substantially reduces the complexity of the diffusion process, thus facilitating room-scale scene

In L3DG, we introduce a novel 3D-VQ-VAE to encode 3D Gaussian Splats. This enables efficient diffusion training, yielding high-quality synthesis of entire rooms and intricate objects! Amazing work lead by Barbara Roessle !

📢Now Hiring: PhD Interns at Meta! 🔍 We're on the hunt for exceptional candidates to intern with our team and work on cutting-edge #DiffusionModels at Meta - GenAI. 💡 Apply Now: metacareers.com/jobs/876309164…

Check out Manuel's great paper on 3D scene generation from a single image by joint shape and pose diffusion!

Fillerbuster: Multi-View Scene Completion for Casual Captures Ethan Weber, Norman Müller, Yash Kant, Vasu Agrawal, Michael Zollhoefer, Angjoo Kanazawa, Christian Richardt tl;dr: latent DiT conditioned on known images and poses (raymaps)->recover unknown content arxiv.org/abs/2502.05175

🚀 Introducing Pippo – our diffusion transformer pre-trained on 3B Human Images and post-trained with 400M high-res studio images! ✨Pippo can generate 1K resolution turnaround video from a single iPhone photo! 🧵👀 Full deep dive thread coming up next!

I'm excited to present "Fillerbuster: Multi-View Scene Completion for Casual Captures"! This is work with my amazing collaborators Norman Müller, Yash Kant, Vasu Agrawal, Michael Zollhoefer, Angjoo Kanazawa, Christian Richardt during my internship at Meta Reality Labs. ethanweber.me/fillerbuster/

Ever wondered how to integrate 3DGS into a pretrained video diffusion model? Check out our new approach, Generative Gaussian Splatting (GGS), that improves 3D-consistency in generated multi-view images. katjaschwarz.github.io/ggs/ arxiv.org/abs/2503.13272

From image(s) to 3D scenes in SECONDS! Bolt3D ⚡️ uses a latent diffusion transformer to generate both image and geometry latents from which we can directly decode 3D Gaussians - no optimization needed.

1/3 Introducing FlowR 🌸: Flowing from Sparse to Dense 3D Reconstructions We learn a direct mapping between incorrect renderings and their corresponding ground-truth images, augmenting scene captures with consistent novel, generated views to improve reconstruction quality.

Check out Tobias' great work leveraging generative priors to improve 3D reconstruction quality!

SOTA 3D Interactive Segmentation We had a lot of fun playing with the possibilities of our real-time 3D segmentation model. Blowing up furniture, rearranging interiors, and of course using the Dune thumpers to turn objects into sand.

Make sure to step by Sherwin’s poster to learn more about camera control for video models!