cv

PhD in Computer Science

• Research on generative models, multimodal understanding, and reinforcement learning.

B.Eng. in Computer Science and Technology

Research Intern

• Prompt engineering and multimodal image understanding for Qwen-Image generation and editing.
• Contributed to efficient video generation and few-step video distillation research.

Research Intern

• Video reasoning and multimodal reinforcement learning.
• Developed Video-Aligner with teacher supervision from the original video branch and consistency rewards for degraded video branches.

Algorithm Intern

• Speech-driven human video generation based on Stable Video Diffusion and mouth-region masking.

Algorithm Intern

• Open-Sora-based autonomous-driving video generation conditioned on BEV scene layouts.

StyleCom

• Training-based cross-domain image composition with style transfer and collage priors.
• Built the OpenImages-Style-600k dataset and led model training and experiments.

Multi-GRPO

• Tree-based trajectory sampling and multi-group advantage estimation for text-to-image reinforcement learning.
• Constructed the OCR-Color-10 text-rendering dataset with explicit color constraints.

Video-Aligner

• Consistency rewards for video reasoning under spatial and temporal degradation.

Compositional Prototypical Networks

• Component prototypes and adaptive fusion for few-shot image classification.

A Few-Shot Image Classification Method Based on Compositional Prototypical Networks

• First inventor; publicly disclosed on March 12, 2024.