About

PhD candidate in Computer Science at the University of Bristol, specialising in efficient AI for image and video. Research spans neural image, video, 3D and 4D scene compression (end-to-end codecs, implicit neural representations, scalable coding), model compression and acceleration (quantization, pruning, knowledge distillation and data-efficient training), generative modelling (flow matching, generative compression), and real-time video models.

2024 —
PhD, Computer Science, University of Bristol. Fully funded by an EPSRC DTP Scholarship.
2021 – 2024
BSc Mathematics and Computer Science, University of Bristol.
Supervisors
Dr Fan Zhang and Prof David Bull
Reviewer
NeurIPS · ICML · ECCV · AAAI · IEEE TCSVT · PCS · ISCAS

News

Oct 2026
My first-author paper S-NVRC accepted at WACV 2027.
Oct 2026
S-NVRC selected for a talk at the 4th AOMedia Research Symposium (4 Nov 2026, Google, Bay Area).
Sep 2026
My first-author paper PerQ accepted at NeurIPS 2026.
Sep 2026
Presented NVRC++ and our Video4Real workshop paper at ECCV 2026 in Malmö. Photos
Aug 2026
My co-first-author paper Toward Real-World Neural Video Representation accepted at the ECCV 2026 Video4Real Workshop.
Jul 2026
My co-first-author paper NVRC++ accepted at ECCV 2026.

Research areas

  • Neural image/video compression
  • Generative modelling for low-level vision
  • 3D Gaussian splatting compression
  • Implicit neural representations
  • Video quality assessment
  • Image super-resolution
  • Dataset condensation
  • Model pruning and knowledge distillation

Publications

WACV · 2027

Scalable Neural Video Representation Compression

“A single embedded bitstream that scales in both bitrate and decoding complexity, outperforming multi-layer VTM by 5.6% BD-rate on UVG.”

Tianhao Peng, Ho Man Kwan, Fan Zhang, Shan Liu, David Bull

arXiv

NeurIPS · 2026

PerQ: Inverse Generative Modeling for Neural Image Compression via Quantization Error Compensation

“A single checkpoint that matches generative codecs in perceptual quality while keeping fidelity competitive with distortion-optimised codecs, across a wide bitrate range.”

Tianhao Peng, Ho Man Kwan, Fan Zhang, Shan Liu, David Bull

Under review · 2026

Neural Reparameterization for 3D Gaussian Splatting Compression

“Generates a 3D Gaussian Splatting scene with small networks instead of storing its Gaussians, so the rate is set by the networks rather than by the number of Gaussians.”

Ho Man Kwan*, Tianhao Peng*, Fan Zhang, David Bull

ECCV Video4Real Workshop · 2026

Toward Real-World Neural Video Representation: Compact, Low-Complexity, and Scalable

Tianhao Peng*, Ho Man Kwan*, Fan Zhang, David Bull

VCIP · 2026

Multi-scale Image Representation Compression

“An overfitted image codec in which every coded component is optimised under a single rate–distortion objective, saving 10.5% BD-rate against VVC.”

Tianhao Peng, Ho Man Kwan, Fan Zhang, Shan Liu, David Bull

arXiv

Preprint · 2026

Advances in Neural Video Compression: A Review and Benchmarking

“The first large-scale unified comparison of conventional and neural video codecs under common test conditions.”

Ge Gao, Chen Feng, Yuxuan Jiang, Tianhao Peng, Ho Man Kwan, Siyue Teng, Chengxi Zeng, Yixuan Li, Changqi Wang, Robbie Hamilton, Fan Zhang, David Bull

Preprint

* Equal contribution.