Over the past year, my lab has been working on fleshing out theory/applications of the Platonic Representation Hypothesis. Today I want to share two new works on this topic: Eliciting higher alignment: arxiv.org/abs/2510.02425 Unpaired rep learning: arxiv.org/abs/2510.08492 1/9

Is the terror reign of redundant scene representations ending? Where VGGT, CUT3R, and other recent models relied on godless redundant outputs (depth+points+pose) without guaranteeing internal prediction consistency, MapAnything and DepthAnything 3 are now heroically pushing back.

3D generation often treats objects as single, monolithic shapes. Our new research, OmniPart, takes a different path: generating objects as assemblies of parts for structural coherence and fine-grained control. And it has been accepted to SIGGRAPH Asia 2025 🎉

A really fun project we've been working on: Real-Time Rendering with JPEG textures! Ended up working way better than we expected (1500+ fps). Will be interesting extending this to modern formats like AVIF or JPEG XL. Paper: arxiv.org/abs/2510.08166 Code: github.com/elias1518693/j…

three years ago, DiT replaced the legacy unet with a transformer-based denoising backbone. we knew the bulky VAEs would be the next to go -- we just waited until we could do it right. today, we introduce Representation Autoencoders (RAE). >> Retire VAEs. Use RAEs. 👇(1/n)

DiT360 High-Fidelity Panoramic Image Generation via Hybrid Training

Pixel-space generative models hit new SOTA with EPG AMAP, Alibaba, NVIDIA & Caltech introduce EPG, a novel two-stage training framework that achieves state-of-the-art pixel-space diffusion (FID 2.04 on ImageNet-256 with 75 NFE) and consistency models (FID 8.82 in 1 step).

MaterialRefGS: Reflective Gaussian Splatting with Multi-view Consistent Material Inference Contributions: • We propose a novel approach to modeling reflections through Gaussian Splatting with multi-view consistent material inference. This includes a multi-view material

A comprehensive, hands-on tutorial on the most recent advancements in robotics 🤟 ...with self-contained explanations of modern techniques for end-to-end robot learning & ready-to-use code examples using LeRobot and Hugging Face. Now available everywhere! 🤗

Super excited to introduce ✨ AnyUp: Universal Feature Upsampling 🔎 Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining. SOTA feature upsampling results while being feature-agnostic at inference time.

Veo is getting a major upgrade. 🚀 We’re rolling out Veo 3.1, our updated video generation model, alongside improved creative controls for filmmakers, storytellers, and developers - many of them with audio. 🧵

🚀 Introducing InstantSfM: Fully Sparse and Parallel Structure-from-Motion. ✅ Python + GPU-optimized implementation, no C++ anymore! ✅ 40× faster than COLMAP with 5K images on single GPU! ✅ Scales beyond 100 images (more than VGGT/VGGSfM can consume)! ✅ Support metric scale.

🎺Meet VIST3A — Text-to-3D by Stitching a Multi-view Reconstruction Network to a Video Generator. ➡️ Paper: arxiv.org/abs/2510.13454 ➡️ Website: gohyojun15.github.io/VIST3A/ Collaboration between ETH & Google with Hyojun Go, Dominik Narnhofer, Goutam Bhat, Federico Tombari, and Konrad Schindler.

Excited to announce a new track of accelerating Generative AI: pi-Flow: Policy-Based Few-Step Generation via Imitation Distillation github.com/Lakonik/piFlow Distill 20B flow models now using just an L2 loss via imitation learning for SOTA diversity and teacher-aligned quality.

[1/N] Excited to introduce "SimULi: Real-Time LiDAR and Camera Simulation with Unscented Transforms." We extend 3DGUT with LiDAR support and render a wide range of sensors 10-20x faster than ray tracing and 1.5-10x faster than prior rasterization work. research.nvidia.com/labs/sil/proje…

what if you could combine diffusion models instantly? You would get exponentially better control (for free!!👀) This is exactly what we do. In ✨ coupled diffusion sampling ✨, diffusion models guide each other. The result? Diverse editing capabilities!

2DGS-R: Revisiting the Normal Consistency Regularization in 2D Gaussian Splatting Contributions: • We conduct extensive experiments to assess the impact of incorporating normal consistency (NC) on the 2DGS attributes. Based on our findings, we propose a hierarchical training

Advances in 4D Representation: Geometry, Motion, and Interaction Abstract (excerpt) Instead of offering an exhaustive enumeration of many works, we take a more selective approach by focusing on representative works to highlight both the desirable properties and ensuing

Mao et al., "PoseCrafter: Extreme Pose Estimation with Hybrid Video Synthesis" While not perfect, video models do an okay job of creating novel views. Use them to "bridge" between extreme views for pose estimation.

Rollouts in the real world are slow and expensive. What if we could rollout trajectories entirely inside a world model (WM)? Introducing 🚀Ctrl-World🚀, a generative manipulation WM that can interact with advanced VLA policy in imagination. 🧵1/6