RAD: A Realistic Multi-View Benchmark for Pose-Agnostic Anomaly Detection

Kaichen Zhou*, Xinhai Chang*, Taewhan Kim*, Jiadong Zhang*, Yang Cao, Chufei Peng, Fangneng Zhan, Hao Zhao, Hao Dong, Kai Ming Ting, Ye Zhu

Paper | Code

• RAD (Realistic Anomaly Detection) is a robot-captured, multi-view dataset designed to address the challenges of pose variation, reflective materials, and viewpoint-dependent defect visibility. It provides a benchmark for pose-agnostic anomaly detection across 13 object categories under uncontrolled lighting conditions.

PAGE-4D: Disentangled Pose and Geometry Estimation for 4D Perception

Kaichen Zhou*, Yuhan Wang*, Grace Chen*, Xinhai Chang, Gaspard Beaudouin, Fangneng Zhan, Paul Pu Liang, Mengyu Wang

Paper | Project | Code

• PAGE-4D is a feedforward 4D perception model that disentangles static and dynamic scene elements to simultaneously achieve accurate camera pose estimation, depth prediction, and point cloud reconstruction in dynamic environments.

A Scalable Whole-body Trajectory Generator for Coordinated Mobile Manipulation

Yida Niu*, Xinhai Chang*, Xin Liu*, Ziyuan Jiao, Yixin Zhu

Paper | Project

• AutoMoMa is a system that efficiently generates high-quality whole-body trajectories using Virtual Kinematic Chain (VKC) modeling and GPU-accelerated motion planning at a rate of 2.5k valid episodes per hour per consumer-level GPU, which generalizes across diverse household layouts, interactive objects, robot morphologies, and manipulation tasks while ensuring physical feasibility and strict constraint satisfaction.

TerraX: Visual Terrain Classification Enhanced by Vision-Language Models

Hongze Li*, Xuchuan Huang*, Xinhai Chang*, Jun Zhou, Huijing Zhao

Paper | Project | Code

• TerraX is a vision-language framework for terrain classification, featuring the TerraData dataset, TerraBench benchmark, and TerraCLIP model.

Neural Surface Reconstruction from Sparse Views Using Epipolar Geometry

Xinhai Chang, Kaichen Zhou

Paper

• EpiS is a generalizable neural surface reconstruction framework that explicitly leverages epipolar geometry. By using an epipolar transformer to fuse multi-view information and incorporating a pretrained monocular depth model for geometry regularization, EpiS achieves state-of-the-art performance in sparse-view surface estimation without per-scene optimization.

SplatMesh: Interactive 3D Segmentation and Editing Using Mesh-Based Gaussian Splatting

Kaichen Zhou*, Lanqing Hong*, Xinhai Chang*, Yingji Zhong, Enze Xie, Hao Dong, Zhihao Li, Yongxin Yang, Zhenguo Li, Wei Zhang

Paper | Code

• SplatMesh is a novel fine-grained interactive 3D segmentation and editing algorithm that integrates 3D Gaussian Splat with a precomputed mesh and could adjust the memory request based on the requirement.