Thesis Project
Developed algorithms and an interactive engine plugin to address core limitations of 3DGS assets lacking Physically-Based Rendering (PBR) properties, relighting capabilities, and LoD strategies. Implemented an inverse rendering algorithm to decouple 3DGS assets into materials and lighting components.
Developed an RL agent (Clipasso-RL) to autonomously generate concise, human-style sketches by learning an adaptive stopping policy. Designed an MDP framework with CLIP embeddings and a hybrid action space.
Developed a multi-agent system for children's creative drawing with an interactive agent, VLM agent, and generative agents. Integrated ComfyUI workflows with Flux-Kontext and LoRA models.
Undergraduate Thesis
Proposed a fiber-level modeling and real-time rendering method for fabrics. Designed a core-fiber and LoD strategy, and a physically-based fiber scattering BSDF.
Provincial-Level Innovation Project
Developed an ML-based strategy for quantitative trading of SSE 50ETF options using LSTM, XGBoost, and CNN-based time-series encoding.