Projects

GVCC: Zero-Shot Generative Video Compression
A pretrained video generative model serves directly as the decoder, and the transmitted bitstream specifies its generation trajectory, enabling zero-shot video compression at ultra-low bitrates. (NeurIPS 2026)
PDF arXiv
Frame Interpolation for Generative Video Coding
Only keyframes are transmitted and the frames in between are regenerated by diffusion-based interpolation, targeting video coding at extremely low bitrates. (Bi-AGMI, IEEE GCCE 2025 Oral Presentation Award; FRS, IEVC 2026)
TSG: Time Step Generating
A universal synthesized deepfake image & video detector that uses a pre-trained diffusion model as a single-pass feature extractor, independent of pretraining models, specific datasets, or sampling algorithms. (ACM MMAsia 2025 Oral)
PDF Code arXiv
Image and Video Coding for Humans and Machines
Scalable and variable-rate coding that serves both machine vision and human viewing, in collaboration with Yui Tatsumi (Watanabe Lab). (IEEE MMSP 2025, IEEE GCCE 2025, IEEE ICCE 2026, MIRU 2025/2026)
Audio-Driven 3D Talking Head
A 3DMM-guided diffusion framework for audio-driven talking head generation based on 3DGS, enabling controllable facial motion and high-fidelity video synthesis.