Haofei Xu徐豪飞

I have broad interests in computer vision and deep learning. I have worked on depth, stereo, optical flow, tracking, feed-forward NeRF/3DGS, and generative models. I am actively exploring new ideas, and I am currently interested in representation learning and generative models for efficient, scalable world models and physical AI. See my full publication list on Google Scholar.

Multi-View 3D Point Tracking

Frano Rajič, Haofei Xu, Marko Mihajlovic, Siyuan Li, Irem Demir, Emircan Gündoğdu, Lei Ke, Sergey Prokudin, Marc Pollefeys, Siyu Tang

ICCV 2025Oral

paper project page interactive results code

The first data-driven multi-view 3D point tracker for tracking arbitrary 3D points across multiple cameras.

PanSplat: 4K Panorama Synthesis with Feed-Forward Gaussian Splatting

Cheng Zhang, Haofei Xu, Qianyi Wu, Camilo Cruz Gambardella, Dinh Phung, Jianfei Cai

CVPR 2025

paper project page code interactive 4K 360 demo

4K panorama synthesis with a single feed-forward inference.

MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images

Yuedong Chen, Haofei Xu, Chuanxia Zheng, Bohan Zhuang, Marc Pollefeys, Andreas Geiger, Tat-Jen Cham, Jianfei Cai

ECCV 2024Oral

paper project page code

A cost volume representation for efficiently predicting 3D Gaussians from sparse multi-view images in a single feed-forward inference.

Unifying Flow, Stereo and Depth Estimation

Haofei Xu, Jing Zhang, Jianfei Cai, Hamid Rezatofighi, Fisher Yu, Dacheng Tao, Andreas Geiger

TPAMI 2023

paper project page slides video(cn) colab demo code

A unified dense correspondence matching formulation enables three motion and 3D perception tasks to be solved with a unified model.

GMFlow: Learning Optical Flow via Global Matching

Haofei Xu, Jing Zhang, Jianfei Cai, Hamid Rezatofighi, Dacheng Tao

CVPR 2022Oral

paper slides video(cn) poster code

Learning cross-view features with a Transformer enables optical flow to be solved by directly comparing feature similarities.

High-Resolution Optical Flow from 1D Attention and Correlation

Haofei Xu, Jiaolong Yang, Jianfei Cai, Juyong Zhang, Xin Tong

ICCV 2021Oral

paper code

Factorizing 2D optical flow with 1D attention and 1D correlation enables 4K resolution optical flow estimation on standard GPUs.

AANet: Adaptive Aggregation Network for Efficient Stereo Matching

Haofei Xu, Juyong Zhang

CVPR 2020

paper code

A sparse points-based cost aggregation method leads to an efficient and accurate stereo matching architecture without any 3D convolutions.

I believe in the power of open source to advance science. Here are some of the key open-source projects stemming from my research and collaborations.