Richard Liu

Thrixel creates game-ready 3D assets that stay fully editable, so you stay in control. Get part-separated models ready for physics and game logic, and switch between editable geometry and photorealistic detail anytime. 🚀 Try Thrixel in beta: thrixel.com/beta

As usual, we have fully open sourced code and an interactive GUI to try these deformations out for yourself. If you are at CVPR, please come chat with us at our poster session! ExHall A&F 555 Poster Session 4 Website: threedle.github.io/dfd/ Code: github.com/threedle/dfd

We explore additional avenues of user control. Fixed points under our framework constrain all points with similar features. Furthermore, users can control the degree of local influence of each control point through our geodesics-based locality weight adjustment.

Very excited to be presenting our highlight paper “Deep Feature Deformation Weights” at #CVPR2026! TLDR: Image features make surprisingly good deformation weights. Our feature distillation is the best in the biz. ExHall A&F 555 Poster Session 4 Website: threedle.github.io/dfd/

📢 Thrilled to present our work Best Segmentation Buddies 🤝 at #CVPR2026 this week! Best Segmentation Buddies is a zero-shot method that matches semantic parts between images in the wild and untextured 3D shapes, even across completely different domains! 🦉✈️ 1/6

Excited to share our #ICCV2025 work Reusing Computation in Text-to-Image Diffusion for Efficient Generation of Image Sets! Our method generates sets of images using significantly less compute than standard diffusion. 📎ddecatur.github.io/hierarchical-d… 1/

We also demonstrate that the curves make effective deformation handles. Our curves bind closely to key semantic features on the original shape surface. Through a simple skinning function, curve deformations can be transferred smoothly and semantically to the shape! /thread

We also show two applications that come out of our optimization design and representation! First is keypoint refinement, in which specific shape details can be included in the abstraction through simply assigning new keypoints to features of interest.

Going to be presenting our oral work WIR3D at @ICCVConference today! Please come check it out! 1:45 pm HST @ Exhibit Hall 3 We will also be #384 at poster session 4 afterwards if you want to chat.