Xiaoming Zhao's photo

I am a Ph.D. student working with Prof. Alexander Schwing at Department of Computer Science, University of Illinois Urbana–Champaign (UIUC).

Earlier, I received B.S. degree in Statistics from University of Science and Technology of China (USTC).

I am passionate about computer vision, generative models, and machine learning, with a broader goal of unifying understanding and generation within vision and beyond.

During my graduate study, I interned at Apple, Meta Reality Labs, and Google, conducting research related to above topics.


Email    /    Google Scholar    /    GitHub    /    CV

Publications

[11]
IllumiNeRF: 3D Relighting Without Inverse Rendering.
Xiaoming Zhao, Pratul P. Srinivasan, Dor Verbin, Keunhong Park, Ricardo Martin Brualla, and Philipp Henzler
Neural Information Processing Systems (NeurIPS), 2024   
[Paper] [Results] [Website] [Leaderboard] [bibtex]   
  
IllumiNeRF provides a simpler approach than traditional inverse rendering for 3D relighting: distilling samples from a single-image relighting diffusion model into a latent-variable NeRF.
[10]
GoMAvatar: Efficient Animatable Human Modeling From Monocular Video Using Gaussians-on-Mesh.
Jing Wen, Xiaoming Zhao, Zhongzheng Ren, Alexander G. Schwing, and Shenlong Wang
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024   
[Paper] [Code] [Website] [bibtex]   

GoMAvatar introduces Gaussians-on-Mesh (GoM) representation for real-time, memory-efficient, and high-quality animatable human modeling.
[9]
NeRFDeformer: NeRF Transformation From a Single View via 3D Scene Flows.
Zhenggang Tang, Zhongzheng Ren, Xiaoming Zhao, Bowen Wen, Jonathan Tremblay, Stan Birchfield, and Alexander G. Schwing
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024   
[Paper] [Code] [Website] [bibtex]   

NeRFDeformer automatically modifies a NeRF representation based on a single RGB-D observation of a non-rigid transformed version of the original scene.
[8]
Pseudo-Generalized Dynamic View Synthesis From a Video.
Xiaoming Zhao, Alex Colburn, Fangchang Ma, Miguel Angel Bautista, Joshua M. Susskind, and Alexander G. Schwing
International Conference on Learning Representations (ICLR), 2024   
[Paper] [Code] [Website] [bibtex]   

PGDVS provides an analysis framework for generalized dynamic view synthesis and finds with consistent depth estimations, scene-specific appearance optimization is NOT required.
[7]
Occupancy Planes for Single-View RGB-D Human Reconstruction.
Xiaoming Zhao, Yuan-Ting Hu, Zhongzheng Ren, and Alexander G. Schwing
AAAI Conference on Artificial Intelligence (AAAI), 2023   
[Paper] [Code] [bibtex]   

OPlanes provides more flexibility than voxel grids and enables to better leverage correlations than per-point classification.
[6]
Generative Multiplane Images: Making a 2D GAN 3D-Aware.
Xiaoming Zhao, Fangchang Ma, David Güera, Zhile Ren, Alexander G. Schwing, and Alex Colburn
European Conference on Computer Vision (ECCV), 2022 (Oral Presentation)   
[Paper] [Code] [Website] [bibtex]   
  
GMPI guarantees to be view-consistent and enables fast training (in less than half a day at a resolution of 10242) and high FPS during inference.
[5]
Initialization and Alignment for Adversarial Texture Optimization.
Xiaoming Zhao, Zhizhen Zhao, and Alexander G. Schwing
European Conference on Computer Vision (ECCV), 2022   
[Paper] [Code] [Website] [bibtex]   

Carefully designed initialization and alignment procedures enable benefiting from both classical and recent learning-based texture optimization techniques.
[4]
Class-agnostic Reconstruction of Dynamic Objects From Videos.
Zhongzheng Ren*, Xiaoming Zhao*, and Alexander G. Schwing
(* denotes equal contribution)
Neural Information Processing Systems (NeurIPS), 2021   
[Paper] [Website] [bibtex]   

REDO enables class-agnostic geometry reconstruction for dynamic objects from RGB-D videos.
[3]
The Surprising Effectiveness of Visual Odometry Techniques for Embodied PointGoal Navigation.
Xiaoming Zhao, Harsh Agrawal, Dhruv Batra, and Alexander G. Schwing
International Conference on Computer Vision (ICCV), 2021   
[Paper] [Code] [Website] [bibtex]   

A well-trained visual odometry module can be a drop-in replacement for GPS and Compass sensor in PointGoal navigation.
[2]
Mitigating Data Scarcity in Protein Binding Prediction Using Meta-Learning.
Yunan Luo*, Jianzhu Ma*, Xiaoming Zhao, Yufeng Su, Yang Liu, Trey Ideker, and Jian Peng
(* denotes equal contribution)
Research in Computational Molecular Biology (RECOMB), 2019   
[Paper] [bibtex]   

Meta-learning and few-shot learning strategy can be utilized to mitigate the data scarcity issue in characterizing the specificity of less-studied kinases for protein-peptide binding prediction.
[1]
Integrating Thermodynamic and Sequence Contexts Improves Protein-RNA Binding Prediction.
Yufeng Su, Yunan Luo, Xiaoming Zhao, Yang Liu, and Jian Peng
PLOS Computational Biology, 2019   
[Paper] [Code] [bibtex]   

A deep learning-based thermodynamic model is introduced for protein-RNA binding prediction.

Workshops

[2]
Learning From Synthesized Demonstrations.
Xiaoming Zhao, Yang Liu, and Jian Peng
International Conference on Machine Learning (ICML) Workshop on Learning in Artificial Open Worlds, 2020   
[Poster] [bibtex]
[1]
Approximation Gradient Error Variance Reduced Optimization.
Wei-Ye Zhao, Yang Liu, Xiaoming Zhao, Jie-Lin Qiu, and Jian Peng
AAAI Conference on Artificial Intelligence (AAAI) Workshop on Reinforcement Learning in Games, 2019   
[Paper] [bibtex]

Slides

Harnessing Data Priors to Mitigate 3D Data Scarcity.   
The slides are almost the same as those for my job talk Harnessing "Dark" Data.   

2024/10: PhD Thesis Defense