Yuanhao Cai

I am currently a 2nd year PhD student in the department of Computer Science, Johns Hopkins University. I am a member of CCVL, advised by Bloomberg Distinguished Professor Dr. Alan Yuille. Previously, I received my MSE and BSE degrees from Tsinghua University in 2023 and 2020. My master advisor is prof. Haoqian Wang. During my study in Tsinghua University, I spent a good time with prof. Yulun Zhang, prof. Xin Yuan, prof. Radu Timofte, and prof. Luc Van Gool.

GitHub  /  Google Scholar  /  ZhiHu

Runner-up Award of NTIRE Low Light Enhancement Challenge at CVPR, 2024

Outstanding Reviewer Award at CVPR, 2024

Outstanding Reviewer Award at ICLR, 2024

Excellent Graduate of Tsinghua University, 2023

Excellent Master Thesis Award, Tsinghua University, 2023

Outstanding Reviewer Award at CVPR, 2023

Top-10 Talented Graduate Student Finalist, Tsinghua University, 2022

National Scholarship, Tsinghua University, 2022

Winner of NTIRE Spectral Reconstruction Challenge at CVPR, 2022

National Scholarship, Tsinghua University, 2021

Winner of COCO Keypoint Detection Challenge at ECCV, 2020

Winner of COCO Keypoint Detection Challenge at ICCV, 2019

Best Paper Award of Joint COCO and Mapillary Recognition Challenge Workshop at ICCV, 2019

2nd Place of AdultSize Drop-In Challenge at RoboCup World Final, 2019

2nd Place of AdultSize Technical Challenge at RoboCup World Final, 2019

3rd Place of AdultSize Soccer Competition at RoboCup World Final, 2019

Learning Progress Scholarship, Tsinghua University, 2019

Science and Technology Innovation Excellence Scholarship, Tsinghua University, 2019

LucidFusion: Generating 3D Gaussians with Arbitrary Unposed Images

A pose-free multi-view 3D reconstruction methods based on Gaussian Splatting

HDR-GS: Efficient High Dynamic Range Novel View Synthesis at 1000x Speed via Gaussian Splatting

The first 3D Gaussian splatting-based method for high dynamic range imaging

R2-Gaussian: Rectifying Radiative Gaussian Splatting for Tomographic Reconstruction

The first 3D Gaussian splatting-based method for CT reconstruction

Radiative Gaussian Splatting for Efficient X-ray Novel View Synthesis

The first 3D Gaussian splatting-based method for X-ray 3D reconstruction

Unveiling Advanced Frequency Disentanglement Paradigm for Low-Light Image Enhancement

A light-weight network for low-light image enhancement

Structure-Aware Sparse-View X-ray 3D Reconstruction

A NeRF algorithm capturing structures for large-scale X-ray 3D reconstruction

Binarized Spectral Compressive Imaging

An Efficient Retinex-based method for Low-light Image Enhancement

Motion-X: A Large-scale 3D Expressive Whole-body Human Motion Dataset

A large-scale human motion benchmark with text description

Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement

An Efficient Retinex-based method for Low-light Image Enhancement

Degradation-Aware Unfolding Half-Shuffle Transformer for Spectral Compressive Imaging

The first Transformer-based deep unfolding method for spectral compressive imaging

Coarse-to-Fine Sparse Transformer for Hyperspectral Image Reconstruction

A novel SOTA Transformer-based method for hyperspectral image reconstruction

Flow-Guided Sparse Transformer for Video Deblurring

The first Transformer-based method for video deblurring

Unsupervised Flow-Aligned Sequence-to-Sequence Learning for Video Restoration

The first Sequence-to-Sequence model for video restoration

MST++: Multi-stage Spectral-wise Transformer for Efficient Spectral Reconstruction

Winner of NTIRE 2022 Challenge on Spectral Reconstruction from RGB. The first Transformer-based method for spectral reconstruction. A baseline and toolbox.

Mask-guided Spectral-wise Transformer for Efficient Hyperspectral Image Reconstruction

The first Transformer-based method for hyperspectral image reconstruction

HDNet: High-resolution Dual-domain Learning for Spectral Compressive Imaging

Dual-domain learning for hyperspectral image reconstruction

RFormer: Transformer-based Generative Adversarial Network for Real Fundus Image Restoration on A New Clinical Benchmark

The first clinical benchmark and Transformer-based method for fundus image restoration

Learning to Generate Realistic Noisy Images via Pixel-level Noise-aware Adversarial Training

A GAN for real noisy image generation

Multi-Scale Selective Feedback Network with Dual Loss for Real Image Denoising

A semi-supervised method for real image denoising

Pseudo 3D Auto-Correlation Network for Real Image Denoising

An efficient architecture for image denoising

Efficient Human Pose Estimation by Learning Deeply Aggregated Representations

An DenseNet-like backbone for efficient human pose estimation

Pyramid Orthogonal Attention Network based on Dual Self-Similarity for Accurate Mr Image Super-Resolution

A novel self-attention mechanism for MR image super-resolution

Learning Delicate Local Representations for Multi-Person Pose Estimation

A multi-stage structure and an attention mechanism for accurate human pose estimation

UDP++

Winner of COCO Keypoint Detection Challenge, 2020

EG^2N: Enhanced Gradient Guiding Network for Single MR Image Super-Resolution

A novel gradient guiding mechanism for MR image super-resolution

Res-Steps-Net for Multi-Person Pose Estimation

Winner of COCO Keypoint Detection Challenge, 2019

Conference: CVPR, ECCV, ICCV, NeurIPS, ICML, ICLR, AAAI, IJCAI, ACM MM, etc.

Journal: TPAMI, IJCV, TIP, TNNLS, Pattern Recognition, etc.