🔥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

All researchers should fight against this. Every week I try to persuade my students that top papers often have few quantitative results. With work that's new, important, and clearly qualitatively different (zero to one!), you don't need quantitative results. Demos not tables!

Modern generative models of images and videos rely on tokenizers. Can we build a state-of-the-art discrete image tokenizer with a diffusion autoencoder? Yes! I’m excited to share FlowMo, with Kyle Hsu, Justin Johnson, Fei-Fei Li, Jiajun Wu. A thread 🧵:

Reminder that the deadline to submit your cool work to the Workshop on 4D vision @ CVPR 2025 is in about a week!!!

🔥Spatial intelligence requires world generation, and now we have the first comprehensive evaluation benchmark📏 for it! Introducing WorldScore: Unifying evaluation for 3D, 4D, and video models on world generation! 🧵1/7 Web: haoyi-duan.github.io/WorldScore/ arxiv: arxiv.org/abs/2504.00983

[1/6] Recent models like DUSt3R generalize well across viewpoints, but performance drops on aerial-ground pairs. At #CVPR2025, we propose AerialMegaDepth (aerial-megadepth.github.io), a hybrid dataset combining mesh renderings with real ground images (MegaDepth) to bridge this gap.

Fresh GPT‑o3 results on our vision‑centric #NaturalBench (NeurIPS’24) benchmark! 🎯 Its new visual chain‑of‑thought—by “zooming in” on details—cracks questions that still stump GPT‑4o. Yet vision reasoning isn’t solved: o3 can still hallucinate even after a full minute of

Just over a month left to submit to this year's Argoverse 2 challenges! Returning from previous years, are our motion forecasting and lidar scene flow challenges. And NEW for this year with a $10k prize pool is our Scenario Mining challenge! 🧵👇

Ever wish YouTube had 3D labels? 🚀Introducing🎥DynPose-100K🎥, an Internet-scale collection of diverse videos annotated with camera pose! Applications include camera-controlled video generation🤩and learned dynamic pose estimation😯 Download: huggingface.co/datasets/nvidi…

Super cool project to have been involved in! Camera motion understanding is far from solved — even top SLAM/SfM and VLM models struggle in the wild. CameraBench pushes the frontier with high-quality annotations and cinematographer-designed taxonomy. VLMs 🤝 SFM next ?😉

What a fun collaboration with Zhiqiu on this summer internship project! Understanding camera motion in videos is extremely challenging, and this CameraBench will be critically important for both video captioning and video generation!

Towards Understanding Camera Motions in Any Video is out on Hugging Face

Supervised learning has held 3D Vision back for too long. Meet RayZer — a self-supervised 3D model trained with zero 3D labels: ❌ No supervision of camera & geometry ✅ Just RGB images And the wild part? RayZer outperforms supervised methods (as 3D labels from COLMAP is noisy)

Compute increases in the last ~decade are insane. The B200 is 1000x faster than the K40 that was state-of-the-art when I started my PhD. We used to train on 1 GPU; now 10K+ is common. Combining these gives a speedup of 10 million since 2013. This explosion led to modern AI.

Last week I passed my thesis proposal, and I'm now officially a Ph.D. candidate! My proposed thesis "Self supervised perception for tactile dexterity" will explore ways to improve dexterous manipulation using tactile reps. Thanks to my committee and everyone that supported me!

Robots need touch for human-like hands to reach the goal of general manipulation. However, approaches today don’t use tactile sensing or use specific architectures per tactile task. Can 1 model improve many tactile tasks? 🌟Introducing Sparsh-skin: tinyurl.com/y935wz5c 1/6

Excited to share our work: Maximizing Confidence Alone Improves Reasoning Humans rely on confidence to learn when answer keys aren’t available (e.g taking an exam). Surprisingly, LLMs can also learn w/o ground-truth answers, simply by reinforcing high-confidence answers via RL!

Check out this shiny new, fast and dynamic web renderer for 3D Gaussian Splats! The things one could do are just mind boggling! So proud of the World Labs team that made this happen, and we are making this open source for everyone!

UFM is a step forward towards solving the top 3 problems of computer vision: Correspondence, Correspondence and Correspondence 🙃 Exciting colab which was led by Yuchen Zhang! 1 year in the making, and lots of engineering and insights uncovered!

UFM: A Simple Path towards Unified Dense Correspondence with Flow Yuchen Zhang, Nikhil Keetha, Chenwei Lyu, Bhuvan Jhamb, Yutian Chen, Yuheng Qiu, Jay Karhade, Shreyas Jha, Yaoyu HU, Deva Ramanan, Sebastian Scherer, Wenshan Wang tl;dr: transformer-based architecture using covisibility

Your bimanual manipulators might need a Robot Neck 🤖🦒 Introducing Vision in Action: Learning Active Perception from Human Demonstrations ViA learns task-specific, active perceptual strategies—such as searching, tracking, and focusing—directly from human demos, enabling robust