Continual Deraining
Compact prompt memories help a single model handle changing rain distributions while preserving earlier knowledge.
Peking University / Computer Vision / Derain
Minghao Liu
I am a Ph.D. student in the School of Computer Science at Peking University, advised by Prof. Jiaying Liu. My research studies image restoration and weather-robust vision, with a focus on making visual systems see clearly through rain and distribution shift. I also keep a small speedcraft notebook for dexterity games and mechanical puzzles, where fast hands meet careful algorithms.
Research base
Wangxuan Institute of Computer Technology, Peking University
Education
B.S. in Computer Science, Turing Class, Peking University
Keywords
Deraining, continual restoration, prompt learning, robust perception
Speedcraft
Former national Butterfly Clicking/CPS record holder; national runner-up in Nine Linked Rings with a 3:04 solve.
TIP 2026
Evolving compact dual prompts
AAAI 2025
Unified image restoration
IJCAI 2023
Continual image deraining
ICME 2023
Efficient video deraining
Research
Restoration models that adapt instead of forgetting.
Efficient Video Restoration
Spatial-temporal distillation transfers stronger video reasoning into lighter deraining models.
Unified Restoration
Unrolled optimization and prompt-driven adaptation provide a bridge between classical priors and modern restoration networks.
Weather-Robust Perception
Domain adaptation improves segmentation and recognition when weather corrupts the visual signal.
Selected Publications
Recent work on deraining and restoration.
Prompting Rain Off: Evolving Compact Dual Prompts for Continual De-Raining
IEEE Transactions on Image Processing, vol. 35, pp. 5266-5280, 2026.
UP-Restorer: When Unrolling Meets Prompts for Unified Image Restoration
A unified restoration framework that combines unrolled optimization with prompt-driven adaptation.
Dual Prompt Learning for Continual Rain Removal from Single Images
A compact prompt-learning scheme for continual single-image deraining under changing data distributions.
Collaborative Spatial-Temporal Distillation for Efficient Video Deraining
A teacher-student framework that distills spatial and temporal structure for efficient video rain removal.
Comprehensive Augmented Domain Adaptation for Image Segmentation Under Rainy Conditions
A weather-aware domain adaptation study for image segmentation in rainy visual conditions.
Honors
Selected academic awards and distinctions.
Outstanding Graduate Student, Wangxuan Institute of Computer Technology, Peking University
Outstanding Graduate, Peking University
Outstanding Undergraduate Research and Outstanding Undergraduate Thesis, Peking University
First-Class Scholarship and Merit Student, Peking University
John Hopcroft Scholarship and Huawei Intelligent Foundation Scholarship
China Mathematical Olympiad Gold Medal and Silver Medal
Beyond Papers
I like research that can be explained, built, played, and shared.
Before and alongside research, I have spent a lot of time mentoring students in mathematics and informatics. I enjoy turning difficult ideas into clean visual explanations, small tools, and systems that are easier to reason about.
I also keep notes on niche skill games: fast clicking, ring puzzles, grid logic, route planning, and other tiny arenas where hand speed and algorithmic thinking become the same habit.
Read the speedcraft notes