Jiafeng Mao
Research Scientist @ CyberAgent AI Lab
About Me
I am currently a research scientist at Cyberagent AI Lab. I worked as a project researcher with Aizawa Lab of The University of Tokyo, Japan.
Recently, I am focusing on agentic visual content understanding and planning
News
- 2026.08: 1 paper is accepted by WACV 2027
- 2026.02: 1 paper is accepted by CVPR workshop AI4CC 2026
- 2026.02: 1 paper is accepted by CVPR 2026
- 2025.11: 1 paper is accepted by AAAI 2026
- 2025.07: 1 paper is accepted by ICCV 2025 Workshop
- 2025.07: 1 paper is accepted by ACM MM 2025
Featured Works
Post-Hoc Domain-Specific Subject Refinement for Generated Commercial Imagery
Qiqi Gao, Jiafeng Mao, Xingchao Yang
- Reference-guided subject correction with scene preservation
- Lightweight LoRA refinement via domain-aligned triplet supervision
Region-Wise Correspondence Prediction between Manga Line Art Images
Yingxuan Li, Jiafeng Mao, Qianru Qiu, Yusuke Matsui
- Transformer-based Manga lineart region matching
- Pseudo & Noisy Label learning
The Lottery Ticket Hypothesis in Denoising: Towards Semantic-Driven Initialization
Jiafeng Mao, Xueting Wang, Kiyoharu Aizawa
- Collecting noise pixel blocks to construct initial noise according to specified layout.
- Comfirming versatility of noise blocks over prompts and images.
Guided Image Synthesis via Initial Image Editing in Diffusion Model
Jiafeng Mao, Xueting Wang, Kiyoharu Aizawa
- Analyzing the generation tendency of intial noise
- initial noise manipulation for generated image repainting and layout-to-image generation
LORE: Latent Optimization for Precise Semantic Control in Rectified Flow-based Image Editing
Liangyang Ouyang, Jiafeng Mao
- Analyzing the generation tendency issue in the image editing task
- Inverted noise optimization for semantic control in RF-based image editing.
Qianru Qiu, Jiafeng Mao, Kento Masui, Xueting Wang
- DiT-based Manga lineart colorization with a reference colorized image
- Hierarchical Attention for Precise region matching