|
Ting-Hsuan Chen I am a first-year Computer Science PhD student at the University of Southern California, advised by Professor Yue Wang. I am honored to be supported by the Viterbi School Taiwan Global Fellowship. I have previously worked as an R&D engineer at Foxconn and also served as a research assistant in Professor Yu-Lun Liu's laboratory at National Yang Ming Chiao Tung University. Currently, I am working at Bosch as a Research Intern focusing on Scene Understanding and Generative AI. |
|
USC 
Bosch 
NYCU 
PSI Lab |
News
|
||||||||||||||||||||||||||||||||||||||
ResearchI'm interested in computer vision, generative AI, and embodied AI. My current research centers on turning generative priors into physical intelligence: using video diffusion and world models to scale up robot data far beyond what teleoperation alone can collect, pushing generated content to respect real-world physics rather than merely look plausible, and repurposing the latent representations learned by large video diffusion and vision-language-action models to improve downstream embodied tasks such as robotic manipulation and humanoid control. |
|
TimePre: Bridging Accuracy, Efficiency, and Stability in Probabilistic Time-Series Forecasting
Lingyu Jiang*, Lingyu Xu*, Peiran Li, Dengzhe Hou, Qianwen Ge, Dingyi Zhuang, Shuo Xing, Wenjing Chen, Xiangbo Gao, Ting-Hsuan Chen, Xueying Zhan, Xin Zhang, Ziming Zhang, Zhengzhong Tu, Michael Zielewski, Kazunori Yamada, Fangzhou Lin, TMLR, 2026 
*: Equal contribution
arXiv TimePre unifies the efficiency of MLP-based models with the distributional flexibility of Multiple Choice Learning for probabilistic time-series forecasting. Its core component, Stabilized Instance Normalization (SIN), corrects channel-wise statistical shifts to resolve catastrophic hypothesis collapse, achieving state-of-the-art accuracy with inference orders of magnitude faster than sampling-based models. |
|
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, CVPR, 2026
*: Equal contribution †: Equal advising
project page / arXiv Pantheon360 generates controllable 360° videos from sparse panoramic inputs. It reconstructs a 3D point cloud cache from the input frames and renders it along user-defined camera trajectories as a geometric scaffold. A video diffusion model then refines this scaffold into photorealistic output, ensuring both geometric consistency and visual quality. |
|
Dino-Diffusion Modular Designs Bridge the Cross-Domain Gap in Autonomous Parking
Zixuan Wu, Hengyuan Zhang, Ting-Hsuan Chen, Yuliang Guo, David Paz, Xinyu Huang, Liu Ren, under review
arXiv / code A modular parking pipeline combining DINOv2 visual foundation models with diffusion-based planning for robust zero-shot transfer across weather and lighting conditions. Achieves 90%+ success rate in cross-domain tests and shows promising sim-to-real transfer. |
|
MoonSim: A Photorealistic Lunar Environment Simulator
Ting-Hsuan Chen*, Henghui Bao*, Ziyu Chen*, Haozhe Lou, Ge Yang, Zhiwen Fan, Marco Pavone, Yue Wang, under review
*: Equal contribution
project page MoonSim is a photo-realistic lunar scene simulator that incorporates Unreal Engine for high-quality lunar images with realistic lighting and shadows and MuJoCo for physics simulation, supporting diverse locomotion and navigation tasks. |
|
NaRCan: Natural Refined Canonical Image with Integration of Diffusion Prior for Video Editing
Ting-Hsuan Chen, Jiewen Chan, Hau-Shiang Shiu, Shih Han Yen, Changhan Yeh, Yu-Lun Liu, NeurIPS, 2024
project page / arXiv / code / demo
|
|
DiffIR2VR-Zero: Zero-Shot Video Restoration with Diffusion-based Image Restoration Models
Changhan Yeh, Chin-Yang Lin, Zhixiang Wang, Chi-Wei Hsiao, Ting-Hsuan Chen, Yu-Lun Liu, arXiv, 2024
project page / arXiv / code / demo
|
Project |
|
DreamMesh
Ting-Hsuan Chen, Cameron Smith, Jiageng Mao, Daniel Wang, github, 2025 A Blender plugin that transforms text or image inputs into complete 3D scenes with rigged objects, realistic AI-generated backgrounds, and intelligent object placement, all powered by generative AI. |
|
GPU-SplineTransformer
Ting-Hsuan Chen github, 2022 My GPU-Optimized SplineTransformer significantly accelerates the conversion of large data arrays into B-spline bases by leveraging GPU power. This innovation outperforms traditional CPU-based solutions, offering enhanced speed and efficiency for your data processing needs. |
Patent |
|
Data Analysis Method, Apparatus, Electronic Device and Storage Medium
Ting-Hsuan Chen TW113141155, 2024 A data analysis method based on feature waveform analysis, improving the accuracy of feature waveform recognition through convolution and segmentation operations. This patent is currently in the confidential stage and is expected to be declassified in 2026. |
Professional Experience |
|
|
PhD Student; MS Student & Research Assistant
Physical Superintelligence Lab, Los Angeles, California, USA Aug 2024 - Present At Physical Superintelligence (PSI) Lab, advised by Professor Yue Wang, I work on video diffusion and world models for embodied intelligence: synthesizing physically consistent robot data at scale, and leveraging the latent representations of large generative models to improve downstream tasks such as robotic manipulation and humanoid control. I completed my MS here in May 2026 and am continuing for my PhD. |
|
|
Scene Understanding/GenAI Research Intern
Bosch Center for Artificial Intelligence, Sunnyvale, California, USA May 2025 - Present At Bosch Center for AI, I work on generative models for scene understanding and autonomous driving. Both Pantheon360 (CVPR 2026) and Dino-Diffusion were developed here. I am also working on two ongoing projects: one improves controllable driving behavior simulation, where current models follow high-level control conditions such as goal and sketch prompts but often do so at the cost of safety and realism; the other explores object-centric video generation, targeting more accurate physics and stronger generalization to unseen objects and scenes. |
|
|
Research Assistant
NYCU Computational Photography Lab, Hsinchu, Taiwan Jan 2024 - June 2024 At the NYCU Computational Photography Lab, my primary research focused on diffusion models. During this period, I successfully published a paper as the first author, which was accepted at NeurIPS 2024. Additionally, I participated in industry-academia collaborations with Nvidia and MediaTek, applying research findings to real-world industry challenges. |
|
R&D Engineer
Foxconn, Hon Hai Precision Industry, Taipei, Taiwan July 2023 - Dec 2023 At Foxconn, I developed the company's first patented ECG waveform recognition system by integrating AI, computer vision, and signal processing techniques. I also mentored new interns and represented the company in various medical conferences. Previously, I built an AI-based data cleansing and classification system, along with essential APIs using Django for medical data processing. |
Awards & Recognitions |
|