Point maps have become a powerful representation for image-based 3D reconstruction. What if we could push point maps even further to tackle 3D registration and assembly?
Introducing Rectified Point Flow (RPF), a generic formulation for point cloud pose estimation.
🧑💻Project: rectified-pointflow.github.io
🔖Paper: arxiv.org/pdf/2506.05282
💻Code: github.com/GradientSpac...
This is a joint work with Tao Sun, Shengyu Huang, Shuran Song and @ir0armeni.bsky.social
🚨 SLAM struggling in dynamic environments? We've been there.
WildGS-SLAM at #CVPR2025, our new monocular RGB SLAM system, tackles dynamic scenes with uncertainty-aware tracking and mapping, resulting to more robust tracking, cleaner maps, and high-quality view synthesis. ⬇️
🌐 wildgs-slam.github.io
Glad to be selected as Outstanding Reviewer for CVPR25!
RPF directly generates the assembled-state point clouds from unposed 3D parts, enabling pose estimation entirely from shape, without correspondence learning, pose regression, or symmetry labels.
This is a joint work with the amazing team: Shengqu Cai, Shengyu Huang, Gordon Wetzstein, Naji Khosravan and @ir0armeni.bsky.social. See you in Vancouver!
Liyuan Zhu
🔔 Want to redesign your apartment and control the style of every piece of furniture? (virtual try-on for 3D scenes).
🎨 Introducing ReStyle3D, a method that transforms your apartment into the design styles as you want! #stylization #SIGGRAPH
Page: restyle3d.github.io
Code: github.com/GradientSpac...
Liyuan Zhu
Leveraging a conditional generative model, our model uncovers and learns shape symmetry and part interchangeability. As a result, it generalizes across categories and datasets, achieving SoTA performance across multiple benchmarks.
🎉 Excited to share our latest work, CrossOver: 3D Scene Cross-Modal Alignment, accepted to #CVPR2025 🌐✨
We learn a unified, modality-agnostic embedding space, enabling seamless scene-level alignment across multiple modalities — no semantic annotations needed!🚀
