Biography
Ziyue Zeng is a Ph.D. student at Kato Laboratory, Waseda University, and a Research Intern at the CG Lab of Huawei Tokyo Da Vinci Research Institute. His research focuses on generative compression, which uses pretrained generative models as decoders to reconstruct images and videos at ultra-low bitrates, and on understanding compressed representations. His recent work, GVCC, a zero-shot video codec built on pretrained video generative models, will appear at NeurIPS 2026.
He received his M.S. from Watanabe Laboratory, Waseda University, where he also worked on diffusion-based frame interpolation and deepfake detection, and his B.S. in Artificial Intelligence from Chongqing University.
- Generative Compression
- Compressed Representation Understanding
- Diffusion-based Deepfake Detection
- Video Generation & Frame Interpolation
- Audio-driven 3D Talking Heads (3DGS)
Ph.D. in Fundamental Science and Engineering (Kato Lab), 2026–present
Waseda University
M.S. in Fundamental Science and Engineering (Watanabe Lab), 2024–2026
Waseda University
B.S. in Artificial Intelligence, 2020–2024
Chongqing University
News
- Sep 2026 — GVCC was accepted to NeurIPS 2026. [arXiv]
- Sep 2026 — Started the Ph.D. program at Kato Laboratory, Waseda University.
- Aug 2026 — GVCCTurbo is out on arXiv. [arXiv]
- Aug 2026 — A co-authored paper received the MIRU 2026 Interactive Presentation Award.
- Jun 2026 — Joined the CG Lab of Huawei Tokyo Da Vinci Research Institute as a Research Intern.
- Mar 2026 — FRS was published at IEVC 2026.
- Dec 2025 — TSG was presented as an oral paper at ACM MMAsia 2025.
- Sep 2025 — Bi-AGMI received the Oral Presentation Award at IEEE GCCE 2025.
Selected Publications
Research Projects
Experience
- Research on generative compression and compressed representation understanding
- Applied diffusion-based frame interpolation to video compression and transmission
- Proposed Bi-AGMI, bidirectional attention-gated keyframe interpolation with diffusion models (IEEE GCCE 2025, Oral Presentation Award)
- Proposed FRS, motion-aware video segmentation for generative video coding (IEVC 2026)
- Proposed GVCC, zero-shot video compression with a pretrained video generative model as the decoder (NeurIPS 2026), and its rate–compute scheduler GVCCTurbo
- Proposed TSG, a universal synthesized image detector built on pretrained diffusion features (ACM MMAsia 2025, Oral)
- Co-authored five papers on scalable image and video coding for humans and machines
- Collaborated on large-motion frame interpolation with Stable Video Diffusion and ControlNet
- Research on evidence theory and information fusion
- Two first-author journal articles (Chaos, Solitons & Fractals; Computational and Applied Mathematics)
Gallery
Contact
Feel free to reach out by email about research or collaboration.
- zengziyue@fuji.waseda.jp
- 3-4-1 Okubo, Shinjuku, Tokyo, 169-8555