M.S.–Ph.D. student · Hefei University of Technology
Making visual intelligence
more trustworthy.
I study how multimodal systems perceive, remember, and reason over visual information — with a focus on building efficient and reliable intelligence for dynamic environments.
Evidence, memory, reasoning.31.82° N · 117.23° E
MULTIMODAL INTELLIGENCE✳STREAMING VIDEO✳VISUAL MEMORY✳MULTIMODAL SAFETY✳MULTIMODAL INTELLIGENCE
SG+
01 / About
Building systems that
keep looking.
I am an M.S.–Ph.D. student in Computer Science at Hefei University of Technology. I am currently advised by Prof. Jiequan Cui, and previously worked under the supervision of Prof. Richang Hong. My broader interest is multimodal intelligence that can efficiently perceive changing environments, retain relevant visual evidence, and reason reliably over long horizons.
Advisors · Jiequan Cui · Richang Hong
02 / Education
The path
so far.
An academic path in computer science, from intelligent systems to multimodal intelligence.
2026 — nowPh.D. · Computer ScienceHefei University of Technology
2024 — 2026M.S. · Computer ScienceHefei University of Technology
2020 — 2024B.Eng. · Intelligent Science & TechnologyHefei University of Technology
03 / Publication
Selected
publication.
My current publication record focuses on image forgery detection and localization. Other directions on this page are presented as research interests and ongoing exploration.
View project code ↗We investigate when semantic priors from vision-language models help — and when they can hurt — forgery detection and localization.
04 / Research interests
Questions I am
exploring.
These are research interests and recent explorations, rather than a list of claimed research specialties. They describe the problems I read, think about, and hope to investigate further.
◌
Multimodal reasoning
How can visual evidence be actively revisited, selected, and used during reasoning instead of being a one-time input?
⌁
Streaming video & memory
Efficient online perception, visual token compression, KV-cache management, and long-term visual memory for dynamic environments.
Read notes ↗◉
Agents & self-distillation
Agentic observation, event-driven memory updates, on-policy self-distillation, and reasoning that remains stable beyond benchmarks.
✦
Multimodal safety
My established publication record is in content safety and image forensics, including robust analysis of generated and manipulated media.
Published work05 / Contact
Let’s talk about
the next question.
I am open to conversations about multimodal intelligence, visual reasoning, content safety, and research collaboration.
shaofengguo@mail.hfut.edu.cn ↗