Jie-Ying Lee 李杰穎

Research Assistant @ NYCU CS
Software Engineer @ Google Pixel Camera

Email: [email protected]

CV / Scholar / LinkedIn / GitHub / X / Threads / Blog

avatar

About Me

I’m a research assistant in Computer Science at National Yang Ming Chiao Tung University, working with Prof. Yu-Lun Liu, and a Software Engineer on Google’s Pixel Camera Team. I work on 3D scene synthesis, generative models for vision, and embodied AI, particularly focusing on Neural Radiance Fields, 3D Gaussian Splatting, vision-language navigation, and on-device perception.

I received my B.S. in Computer Science from National Yang Ming Chiao Tung University, with an exchange semester at ETH Zurich. My industry experience includes internships at Google (Pixel Camera Team), Microsoft, and Appier.

I am actively seeking research collaborations. If you are interested in working with me, don’t hesitate to reach out.

Google Google SWE (2025 - Present)
NYCU NYCU Research Assistant (2025 - Present) B.S in Computer Science (2021 - 2025)
ETH Zurich ETH Zurich Exchange Student (2024 - 2025)

News

Publications

Skyfall-GS: Synthesizing Immersive 3D Urban Scenes from Satellite Imagery teaser image ▶ Hover / Tap
Skyfall-GS: Synthesizing Immersive 3D Urban Scenes from Satellite Imagery
Jie-Ying Lee, Yi-Ruei Liu, Shr-Ruei Tsai, Wei-Cheng Chang, Chung-Ho Wu, Jiewen Chan, Zhenjun Zhao, Chieh Hubert Lin, Yu-Lun Liu
ECCV, 2026
project page / arXiv / pdf / code★ 941 / video

We present Skyfall-GS, a framework that synthesizes photorealistic, city-block scale 3D urban scenes from satellite imagery using diffusion models, eliminating the need for expensive 3D scanning and manual annotation while enabling real-time exploration.

🎬 Featured in This AI Just Fixed Google Earth’s Biggest Problem by Bilawal Sidhu — 230K+ views on YouTube.

BRDFusion: Physics Meets Generation for Urban Scene Inverse Rendering teaser image ▶ Hover / Tap
BRDFusion: Physics Meets Generation for Urban Scene Inverse Rendering
Yi-Ruei Liu, Jie-Ying Lee, Zheng-Hui Huang, Yu-Lun Liu†, Chih-Hao Lin† (†Corresponding Author)
arXiv, 2026
project page / arXiv / pdf / code★ 43
We present BRDFusion, an inverse rendering framework for urban scenes that unifies physics-based rendering with generative diffusion priors to recover explicit material and lighting properties from video, enabling high-quality relighting and simulation.
Pantheon360: Taming Digital Twin Generation via 3D-Aware 360° Video Diffusion teaser image ▶ Hover / Tap
Pantheon360: Taming Digital Twin Generation via 3D-Aware 360° Video Diffusion
Ting-Hsuan Chen*, Ying-Huan Chen*, Tao Tu, Jie-Ying Lee, Cho-Ying Wu, Fangzhou Lin, Hengyuan Zhang, David Paz, Xinyu Huang, Yuliang Guo†, Yu-Lun Liu†, Yue Wang†, Liu Ren (*Equal Contribution, †Corresponding Author)
CVPR, 2026
project page / arXiv
We present Pantheon360, a controllable 360° video diffusion framework for digital twin generation that synthesizes high-fidelity videos from sparse 360° inputs, letting the diffusion model focus on photorealistic texture refinement while a 3D Cache enforces global geometric consistency.
LightsOut: Diffusion-based Outpainting for Enhanced Lens Flare Removal teaser image ▶ Hover / Tap LightsOut: Diffusion-based Outpainting for Enhanced Lens Flare Removal hover image
LightsOut: Diffusion-based Outpainting for Enhanced Lens Flare Removal
Shr-Ruei Tsai, Wei-Cheng Chang, Jie-Ying Lee, Chih-Hai Su, Yu-Lun Liu
ICCV, 2025
project page / arXiv / code★ 31 / video / demo
We present LightsOut, a diffusion-based outpainting framework that enhances lens flare removal by reconstructing off-frame light sources. Our approach combines a multitask regression module with LoRA fine-tuned diffusion models to produce realistic and physically consistent results.
See, Point, Fly: A Learning-Free VLM Framework for Universal Unmanned Aerial Navigation teaser image ▶ Hover / Tap
See, Point, Fly: A Learning-Free VLM Framework for Universal Unmanned Aerial Navigation
Chih-Yao Hu*, Yang-Sen Lin*, Yuna Lee, Chih-Hai Su, Jie-Ying Lee, Shr-Ruei Tsai, Chin-Yang Lin, Kuan-Wen Chen, Tsung-Wei Ke, Yu-Lun Liu (*Equal Contribution)
CoRL, 2025
project page / arXiv / code★ 155 / video
We present See, Point, Fly (SPF), a training-free framework for aerial vision-and-language navigation. By leveraging vision-language models and reformulating navigation as a 2D spatial grounding task, SPF enables universal unmanned aerial navigation without task-specific training.
AuraFusion360: Augmented Unseen Region Alignment for Reference-based 360° Unbounded Scene Inpainting teaser image ▶ Hover / Tap
AuraFusion360: Augmented Unseen Region Alignment for Reference-based 360° Unbounded Scene Inpainting
Chung-Ho Wu*, Yang-Jung Chen*, Ying-Huan Chen, Jie-Ying Lee, Bo-Hsu Ke, Chun-Wei Tuan Mu, Yi-Chuan Huang, Chin-Yang Lin, Min-Hung Chen, Yen-Yu Lin, Yu-Lun Liu (*Equal Contribution)
CVPR, 2025
project page / arXiv / code★ 83 / video
We introduce AuraFusion360, a reference-based 360° scene inpainting method with three key innovations: depth-aware occlusion identification, Adaptive Guided Depth Diffusion for zero-shot point placement, and SDEdit-based enhancement for multi-view coherence.
SpectroMotion: Dynamic 3D Reconstruction of Specular Scenes teaser image ▶ Hover / Tap
SpectroMotion: Dynamic 3D Reconstruction of Specular Scenes
Cheng-De Fan, Chen-Wei Chang, Yi-Ruei Liu, Jie-Ying Lee, Jiun-Long Huang, Yu-Chee Tseng, Yu-Lun Liu
CVPR, 2025
project page / arXiv / code★ 8 / video
We present SpectroMotion, the first 3D Gaussian Splatting method capable of reconstructing photorealistic dynamic specular scenes. By combining 3DGS with physically-based rendering and deformation fields, we achieve high-quality synthesis of challenging real-world dynamic reflective surfaces.
BoostMVSNeRFs: Boosting MVS-based NeRFs to Generalizable View Synthesis in Large-scale Scenes teaser image ▶ Hover / Tap
BoostMVSNeRFs: Boosting MVS-based NeRFs to Generalizable View Synthesis in Large-scale Scenes
Chih-Hai Su*, Chih-Yao Hu*, Shr-Ruei Tsai*, Jie-Ying Lee*, Chin-Yang Lin, Yu-Lun Liu (*Equal Contribution)
SIGGRAPH, 2024
project page / arXiv / code★ 28 / video
We present BoostMVSNeRFs, a method that enhances rendering quality for MVS-based NeRFs in large-scale scenes. Our approach addresses key limitations including restricted viewport coverage and artifacts from limited input views, enabling generalizable view synthesis in complex environments.

Projects

Physically-based Renderer based on Nori 2 teaser image
Physically-based Renderer based on Nori 2
Jie-Ying Lee*, Wei Wang* (*Equal Contribution)
Computer Graphics, ETH Zurich, 2024
blog post / report / slides
Built a complete physically-based renderer in C++ on top of the Nori 2 educational framework: path tracing with multiple importance sampling, Disney BSDF, image and procedural textures, a realistic camera model, and Intel Open Image Denoise integration. Our competition entry In Between — a dawn view from a waterfall cave inspired by Seljalandsfoss, Iceland — renders the clouds and waterfall mist as heterogeneous participating media from NanoVDB volumes, combining equiangular sampling with the Henyey-Greenstein phase function under environment-map lighting. Selected as a finalist of the course rendering competition.
Operating System Capstone: Bare-metal Kernel on Raspberry Pi teaser image
Operating System Capstone: Bare-metal Kernel on Raspberry Pi
Jie-Ying Lee
Operating System Capstone, NYCU, 2024
code
Implemented a bare-metal OS kernel in C and assembly for Raspberry Pi 3B+ (ARMv8-A, AArch64) across eight labs: UART shell and mailbox interface, a self-hosted UART bootloader with device tree parsing, exception and interrupt handling, a buddy-system memory allocator, threads and user processes with system calls and signals, virtual memory, and a virtual file system with tmpfs, initramfs, and device file systems.
P Language Compiler teaser image
P Language Compiler
Jie-Ying Lee
Compiler Design, NYCU, 2024
code
Built a complete compiler in C++ for P, a Pascal-like language, using flex and bison: scanner, parser, AST construction, semantic analysis with symbol tables, and RISC-V code generation executed on the Spike simulator.

Service

Misc.

Beyond research, I’m passionate about staying active through badminton and hip-hop dance. I also enjoy capturing moments through photography.

Music-wise, I’m into Taiwanese indie and hip-hop, frequently listening to artists like Gummy B, 草東沒有派對 (No Party For Cao Dong), and 國蛋 GorDoN.