Yingtao Li

Yingtao Li李英韬 李英韬

Undergraduate Student, School of Artificial Intelligence
Dalian University of Technology 大连理工大学 人工智能学院
本科生

I am a junior undergraduate majoring in AI at DUT (GPA 3.9/5.0, rank 11/312). My research interests focus on 3D Gaussian Splatting, Computer Vision, and Image Style Transfer. I have one paper accepted at CVPR 2026 and one under review at IEEE T-CSVT. 我是大连理工大学人工智能专业的本科三年级学生（GPA 3.9/5.0，专业排名 11/312）。我的研究兴趣集中在3D高斯溅射（3DGS）、计算机视觉和图像风格迁移。目前我有一篇论文被 CVPR 2026 接收，另一篇在 IEEE T-CSVT 审稿中。

Email邮箱 GitHub CSDN

News最新动态

  • 2026 One paper accepted at CVPR 2026.一篇关于风格迁移的论文被 CVPR 2026 接收。
  • 2025 One paper submitted to IEEE T-CSVT.一篇关于 3DGS 的论文投递至 IEEE T-CSVT 审稿中。
  • 2025 Awarded National Scholarship.荣获 国家奖学金。

Publications发表论文

CVPR 2026CVPR 2026

Style-GRPO: Semantic-Aware Preference Optimization for Image Style Transfer Guided by Reward Modeling

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026 — Accepted IEEE/CVF 计算机视觉与模式识别会议 (CVPR), 2026 — 已接收

A reward-model-guided preference optimization framework for semantically coherent image style transfer. 一种基于奖励模型引导的语义感知图像风格迁移偏好优化框架。

Under Review在投

Breaking the Vicious Cycle: Coherent 3D Gaussian Splatting from Sparse and Motion-Blurred Views

IEEE Transactions on Circuits and Systems for Video Technology (T-CSVT) — Under Review IEEE 视频技术电路与系统汇刊 (T-CSVT) — 审稿中

Addressing sparse-view and motion-blur degradation in 3D Gaussian Splatting for coherent novel-view synthesis. 解决3DGS在稀疏视角和运动模糊下的退化问题，实现连贯的新视角合成。

Research研究方向

Ongoing进行中

3D EventVGGT / StreamVGGT

Fusing RGB and Event camera data for robust 3D reconstruction. StreamVGGT employs Cross-Attention with RGB tokens as Query and Event tokens as Key/Value, supporting real and simulated domains. Achieves ~20% improvement over baselines on 15 SIM/REAL benchmarks.

融合RGB和事件相机数据进行鲁棒的三维重建。StreamVGGT采用交叉注意力机制，以RGB token为Query，Event token为Key/Value，支持真实和模拟域。在15个基准测试中比基线提升约20%。

Event Camera Cross-Attention 3D Reconstruction
Ongoing进行中

Differentiable Compatibility Prior (DCP)

Exploring differentiable priors for improving compatibility and consistency in 3D vision tasks.

探索可微先验以改善3D视觉任务中的兼容性和一致性。

Differentiable Rendering 3D Vision
Published已发表

Style-GRPO

Semantic-aware preference optimization for image style transfer using reward modeling. Accepted at CVPR 2026.

基于奖励模型的语义感知图像风格迁移偏好优化。被 CVPR 2026 接收。

Style Transfer GRPO Reward Modeling
Under Review在投

Coherent 3DGS from Degraded Views

Breaking the vicious cycle of sparse and motion-blurred inputs in 3D Gaussian Splatting. Submitted to IEEE T-CSVT.

打破3DGS中稀疏和运动模糊输入的恶性循环。已投递至 IEEE T-CSVT。

3D Gaussian Splatting Motion Blur Sparse View

Honors & Skills荣誉与技能

Awards & Honors荣誉奖项

  • 2025 National Scholarship国家奖学金
  • 2025 GPA 3.9/5.0, Rank 11/312GPA 3.9/5.0, 专业排名 11/312
  • 2026 CVPR 2026 Paper AcceptedCVPR 2026 论文接收

Education教育背景

Dalian University of Technology大连理工大学 2024 – Present2024 – 至今
B.Eng. in Artificial Intelligence人工智能专业 工学学士
GPA: 3.9/5.0GPA: 3.9/5.0 | Rank: 11/312排名: 11/312
Key courses: Python (98), Linear Algebra (97), Probability & Statistics (94), Signal Processing (92) 核心课程: Python (98), 线性代数 (97), 概率论与数理统计 (94), 信号与系统 (92)

Technical Skills技术栈

Languages编程语言

Python C/C++ MATLAB LaTeX

Frameworks深度学习框架

PyTorch Hydra Accelerate DDP / FSDP

Architectures模型架构

Transformer Cross-Attention 3D Gaussian Splatting Voxelization

Tools开发工具

Linux Git / GitHub tmux Docker

Contact联系方式

Email邮箱 yingtaoli38@gmail.com
Phone电话 135-9176-0428
GitHub @Passwerob
CSDN Passwerob