Xingyi Yang

DirMoE: Dirichlet-Routed Mixture of Experts

Visual Geometry Transformer in the Wild: Distractor-Free 3D Reconstruction

SparseD: Sparse Attention for Diffusion Language Models

Revisit Model Adaptation from Parameters to Features

Epipolar Prompt: A Simple Baseline for Motion Segmentation

Don't Forget the Nonlinearity: Unlocking Activation Functions in Efficient Fine-Tuning

Disentangling Token Dependencies for Efficient Decoding in Diffusion Language Models

Image Editing As Programs with Diffusion Models

GraphBridge: Towards Arbitrary Transfer Learning in GNNs

TreeSplat: Mergeable Tree for Deformable Gaussian Splatting

Test3R: Learning to Reconstruct 3D at Test Time

Kolmogorov-Arnold Transformer

Compositional Video Generation as Flow Equalization

Mixture of Experts Made Intrinsically Interpretable

1000+ FPS 4D Gaussian Splatting for Dynamic Scene Rendering

Hash3D: Training-free Acceleration for 3D Generation

Video-Infinity: Distributed Long Video Generation