Chen Change Loy

4RC introduces a unified, fully feed-forward framework for monocular 4D reconstruction that encodes an entire video once and then flexibly queries dense 3D geometry and motion for any frame at any timestamp. By factorizing scene structure into base geometry and time-dependent displacements, it achieves accurate, efficient, and state-of-the-art performance across diverse 4D reconstruction tasks. Paper: arxiv.org/pdf/2602.10094 Project Page: yihangluo.com/projects/4RC/

We dropped our dataset last week. Just 10M frames of real hands touching real objects. 5 days later → 1.66M downloads Today → 2.12M and climbing Status → Most trending on @huggingface Turns out the entire Physical AI community has been starving for real-world data. Foundation models dream in pixels and tokens. But they've never felt weight. Never opened a door. Never experienced actual physics. That changes now. #PhysicalAI #EmbodiedAI #Robotics #AI #HuggingFace

Ropedia Xperience-10M is out on Hugging Face a large-scale egocentric multimodal dataset of human experience for embodied AI, robotics, world models, and spatial intelligence It contains 10 million experiences (interaction) and 10,000 hours of synchronized first-person recordings with six video streams, audio, stereo depth, camera pose, hand mocap, full-body mocap, IMU, and hierarchical language annotations dataset: huggingface.co/datasets/roped…

Hot take: "More data" won't fix AI. If the data is static, you just get a bigger model that still doesn't understand the real world. Intelligence lives in sequences. Not snapshots. That's why we're building systems that act in the physical world — long-horizon sequential tasks, dozens of precise actions, zero room for error. End to end. Full demo here → x.com/ropedia_ai/sta… #EmbodiedAI #Robotics #LongHorizonPlanning #AI #FutureOfAI

MatAnyone 2 is out on Hugging Face Scaling Video Matting via a Learned Quality Evaluator paper: huggingface.co/papers/2512.11… app: huggingface.co/spaces/Peiqing…

MatAnyone2, the next step in scalable video matting. Compared with MatAnyone, which leveraged consistent memory propagation for stable target-assigned video matting, MatAnyone2 introduces a Matting Quality Evaluator (MQE) that provides pixel-wise feedback on semantic and boundary quality. Key advances: - Learned MQE enables fine-grained supervision and data filtering - VMReal dataset with 28K real-world clips / 2.4M frames for scalable training - Reference-frame training to handle long-range temporal variations - SOTA performance on synthetic and real-world benchmarks This pushes video matting closer to practical deployment in real-world pipelines. Paper: arxiv.org/abs/2512.11782 Project: pq-yang.github.io/projects/MatAn… Code: github.com/pq-yang/MatAny… Congrats to @peiqing001 and the team for the excellent work! 🎉

@Poiex @DRAVEN_ALG @TheYihangLuo @KangLiao929 I am a late comer 😂

@ccloy @DRAVEN_ALG @TheYihangLuo @KangLiao929 happy to see you started working on VLAs too 😉

VLANeXt systematically studies what actually matters for VLA performance: policy heads, meta-queries, action chunking, flow matching, temporal history, multi-view, and proprioception, and shows that a simple time-series forecasting perspective improves action prediction. 📄 Paper: arxiv.org/abs/2602.18532 💻 Code: github.com/DravenALG/VLAN… If you’re working on VLAs, also check out this curated list of resources: 🔎 github.com/DravenALG/awes…

VLANeXt Recipes for Building Strong VLA Models huggingface.co/papers/2602.18…

Anthropic just introduced Cowork, an AI assistant that runs on your computer — but they can see your data ⚠️⚠️⚠️ We have a secure alternative: 🍫 Local Cocoa — a fully 💻on-device, 🌍open-source AI assistant that runs completely offline. Code: github.com/synvo-ai/Local…

🎯Local Cocoa: Your Personal AI File Assistant, Fully Local💻 @synvoAI open-sourced 🍫Local Cocoa🍫, which turns millions of local files with various types (e.g., text, video, audio) into a 🧠structural multimodal memory🧠 for agentic contextual intelligence - Repo: github.com/synvo-ai/Local… 💡Each file turns into memory. Memories form context. Context sparks insight. Insight powers action.

🚀 @synvoAI We’re excited to open-source Local-Cocoa, an on-device AI assistant that turns your files into semantic memory while keeping everything private. Built in collaboration with @MMLabNTU, and we warmly welcome contributors to help push it further! 🔧✨ 👉 github.com/synvo-ai/Local…

We introduce Log-linear Sparse Attention (LLSA), a trainable sparse attention mechanism that reduces attention complexity from O(N²) to O(N log N). 📄 Paper: arxiv.org/abs/2512.16615 💻 Code: github.com/SingleZombie/L…

Encouraging to see this. As VLAs scale, human video and robot data begin to align, making human experience a usable training signal. This is exactly what I’ve believed in, and what we’ve been building at @ropedia_ai .

⚡️Thrilled to introduce @ropedia_ai, where we aim to structure human Xperience (observations + interactions) for world models, robotics foundation models and interactive intelligence! 🔥Check out our blog for more: ropedia.com/blog/20251216_… 🎁Don’t forget to join our waitlist for early access of HOMIE: docs.google.com/forms/d/e/1FAI…

Big congrats to @ropedia_ai, a spin-off from @MMLabNTU, pushing the frontier of how we capture and interact with the physical 4D world 🚀 Another lab spin-off, @synvoAI, is reimagining contextual memory for next-gen AI systems.

Congrats to @synvoAI and SBG!

NeurIPS 2025 papers per 1 Million People 1. Singapore – 64.51 2. Switzerland – 22.13 3. Israel – 11.17 4. UAE – 9.47 5. UK – 7.50 6. US – 7.44 7. Denmark – 7.37 8. Australia – 7.31 9. Canada – 6.93 10. South Korea – 5.78

ICLR has placed OpenReview in a difficult position, so I want to offer a few words about the OpenReview team working behind the scenes. OpenReview has long been operated at UMass Amherst as a non-profit organization founded by Andrew McCallum. Each year, Andrew must raise more than $2 million to support a 20-person team that provides essential infrastructure for most major conferences. I once asked Andrew what might have been a naïve question: whether he had considered developing a business model for OpenReview, given its prominence and the seemingly obvious opportunities. He pushed back, explaining that everything he has done for OpenReview is driven by a commitment to serve and strengthen the academic community. He is willing to devote significant personal effort to ensure the platform remains freely accessible to all. We should not blame such a brilliant and dedicated team for an accidental issue. Otherwise, fewer people would be willing to shoulder this kind of responsibility in the future. Deep respect to the OpenReview team! I’m grateful for their work and happy to support in any way!

🚀Thrilled to announce that Synvo AI has joined the NVIDIA Inception program, a significant step forward in accelerating our mission to build context-aware AI solutions! #NVIDIAInception #DeepTech #Entrepreneurship