Siyuan Fan

I am working on AINR/ISP @Axera in Hangzhou.

Research

I'm interested in computational photography, image signal processing, real-world image and video denoising, deep learning, and efficient AI systems for edge devices. My work focuses on making vision algorithms robust to challenging sensor noise and practical deployment constraints.

Education

Beihang University — M.E. in Optical Engineering, 2020.9–2023.1

Changchun University of Science and Technology — B.E. in Optoelectronic Information Science and Engineering, 2016.9–2020.6

Experience

AI-ISP Algorithm Engineer, Axera — Hangzhou, 2023.2–Present

Built an AI2DNR RAW-domain paired-data pipeline covering noise synthesis, distillation, and validation for multiple sensor processes.
Designed FPN correction and SNR-guidance calibration for temporally correlated noise in multi-frame denoising.
Developed Hybrid VST for extremely dark scenes, improving robustness to black-level drift and sensor calibration errors.
Built diffusion-inspired controllable denoising data augmentation and contributed to LoRA-based strength control.
Developed parts of an AIISP simulator, including ONNX-to-CUDA operators and tile-based execution for large images.

Selected Publications & Patents

SAT: Shift Alignment Transformer for Video Denoising Without Flow Estimation. Scientific Reports, 2026.