Xiangyu Xu

I am a full Professor at Xi'an Jiaotong University. I work on computer vision and machine learning. I got my Bachelor degrees from Tsinghua University and Peking University. I did my PhD at Tsinghua University and my Postdoc at Carnegie Mellon University.

I am looking for PhD, Master, and Undergraduate students and Postdocs to join my lab. Please drop me an email if you are interested.

Email  /  Bio  /  Google Scholar

I'm interested in computer vision, machine learning, and image processing.

A new framework for drag editing with diffusion models.

We combine Level of Detail (LoD) with NeRF.

Learnable filters for image deblurring.

We propose a very simple Transformer for 3D human shape and pose estimation from a single image, which achieves SOTA accuracy with high efficiency in parameter and computation.

Text-to-3D generation without per-prompt training, taking only under a second.

A progressive strategy that learns text-to-3D in a coarse-to-fine manner

We achieve new SOTA for single-view 3D reconstruction on CO3D.

We present the first solution for sewing pattern reconstruction from a single image.

We explore privacy-preserving action recognition from a spatio-temporal perspective.

We propose a cylinder-style convolution for completing panoramic views.

We develop a lighter version of GLEAN.

We present an SMPL-based NeRF for multi-view human synthesis, which is efficient and generalizable to unseen subjects and sparse views.

We present the first Transformer architecture for video frame interpolation.

Several useful techniques for real-world video super-resolution.

Winner algorithm for video enhancement challenges in NTIRE 2021.

We present a Transformer for 3D human texture estimation from a single image.

We extend RSC-Net to videos and propose a human texture estimation model.

We use StyleGAN for image super-resolution.

For the first time, super-resolution is applied to human-computer interaction.

We improve the RawSR architecture and extend it to image dehazing and depth upsampling.

We propose a new algorithm for 3D human shape and pose estimation that is robust to low-resolution input.

We propose spatio-temporal deformable convolution kernels for image and video denoising.

We propose deep Robust PCA for joint image filtering.

We present the first nonlinear motion model for image interpolation. The method won the Champion of AIM 2019 video interpolation challenge.

We present the first neural network for raw image based super-resolution. It works very well on real-world input.

We present a hybrid structure of CNN and RNN for low-light image enhancement.

We present a CRF-based pipeline and a deep neural filter for rendering shallow depth-of-field effect with monocular camera.

We propose to incorporate relative features, i.e., affinity, for monocular depth estimation.

We present a deep neural network to extract salient edges for blind image deblurring.

We exploit a neural network to perform haze removal for videos.

We present the first deep neural network for blind image super-resolution.

Area Chair: CVPR 2023, BMVC 2023, CVPR 2024, ICLR 2024, ECCV 2024
Session Chair: SIGGRAPH Asia 2023
Senior Program Committee: IJCAI 2021, AAAI 2023, AAAI 2024

Stefan Lionar, PhD (Sea AI Lab and NUS, 2023-present)
Ming Li, Intern (Sea AI Lab, 2023)
Hao Chen, Master (CMU, 2019-2020), Now: PhD at CMU
Muchen Li, Intern (SenseTime, 2018-2019), Now: PhD at UBC
Yongrui Ma, Intern (SenseTime, 2018-2019), Now: PhD at CUHK

Website template from here