I am a Computer Science Ph.D. student at the University of Massachusetts (UMass) Amherst, advised by Professor Ramesh K. Sitaraman and Professor Mohammad Hajiesmaili. Before joining UMass, I received a M.S. degree in Computer Science from National Taiwan University (NTU), where I worked with Professor Winston Hsu.
My research sits at the intersection of systems and machine learning, focusing on practical volumetric video streaming systems. In particular, I develop neural-enhanced content-distribution frameworks that capture, compress, deliver, and render 3D visual media. This entails minimizing both the size of the 3D representations and the ML inference latency, all while preserving high visual quality. Currently, I am exploring 1) real-time encoding for point cloud sequences and 3D Gaussian Splatting by leveraging standard video codecs to compress re-organized 3D data, and 2) efficient serving of text-to-image diffusion models by enabling caching and model scaling.
Prior to my Ph.D. work, I focused on few-shot learning in computer vision, with applications that enhance the generalizability and adaptability of object detectors.
Selected Publications
VM-NIF: Real-time View Morphing via Neural Implicit Function
Tung-I Chen, D.-Y. Lee, G.-M. Su, Mohammad Hajiesmaili, Ramesh K. Sitaraman
Under Submission, Patent Filed by Dolby
Dynamic Density-Aware Active Domain Adaptation for Semantic Segmentation
Tsung-Han Wu, Y.-S. Liou, S.-J. Yuan, H.-Y. Lee, Tung-I Chen, K.-C. Huang, Winston H Hsu
Accepted by European Conference on Computer Vision (ECCV), 2022
[Paper]
Dual-Awareness Attention for Few-Shot Object Detection
Tung-I Chen, Y.-C. Liu, H.-T. Su, Y.-C. Chang, Y.-H. Lin, J.-F. Yeh, Winston Hsu
Accepted by IEEE Transactions on Multimedia (TMM), 2021
[Paper, Code]
News
[September 2024] Our work “VM-NIF: Real-time View Morphing via Neural Implicit Function” has been filed as a patent by Dolby!
[April 2024] Received PhD research internship from Dolby Laboratories