Projects

A deep unfolding framework introducing stage-specific LoRA adapters to dynamically modulate denoising behavior across unfolding stages. Achieves N× parameter reduction for N-stage networks while maintaining competitive performance across multiple image restoration tasks.
Keywords: Deep Unfolding, Low-Rank Adaptation, Image Restoration

A framework for spectral compressive imaging reconstruction that leverages diffusion models to recover high-frequency details from a single compressed measurement, addressing the challenges of limited MSI training data and high computational demands.
Keywords: Spectral Imaging, Diffusion Models, Image Reconstruction

[image]

A local media management tool for photographers featuring intelligent grouping via HSV histogram similarity, aesthetic scoring using ZJU LAR-IQA algorithm (ONNX Runtime / DirectML acceleration), and one-click export. Performance: ~3ms/frame thumbnail generation, ~1s/frame quality assessment.
Keywords: Image Quality Assessment, GPU Acceleration, Media Management

[image]

A cross-modal generation pipeline using CLIP embeddings for music-to-image translation. Designed a unified framework that bridges audio understanding and visual generation.
Keywords: Cross-Modal Generation, CLIP, Music-to-Image