publications

2026

1. SIGGRAPH

   LUCID: Learning Unified Control for Image Deflaring and Exposure Mastery in Nighttime Photography

   Tingyu Yang, Yuan Cheng, and Xiaoyun Yuan

   In ACM SIGGRAPH 2026 Conference Papers, 2026

   Abs arXiv Code Hugging Face Website

   We present LUCID, a unified controllable image restoration framework for nighttime photography that jointly handles lens flare removal and exposure adjustment. By combining flare decoupling with Diffusion CFG control, LUCID achieves continuous brightness adjustment through mixed training and supports text-guided light source preservation, enabling multi-mode training and multi-effect inference within a single model. The framework also leverages multi-exposure scene reconstruction for high-quality nighttime HDR synthesis, achieving state-of-the-art IQA metrics on large-scale real-world nighttime datasets.

2. AAAI

   SODiff: Semantic-Oriented Diffusion Model for JPEG Compression Artifacts Removal

   Tingyu Yang^*, Jue Gong^*, Jinpei Guo, and 3 more authors

   In Proceedings of the AAAI Conference on Artificial Intelligence, 2026

   Abs arXiv

   We propose SODiff, an image restoration method for JPEG compression artifact removal that aligns image features with semantic guidance to better exploit pretrained generative priors. By bridging compressed image features with semantic information, SODiff improves texture reconstruction naturalness and overall perceptual quality under high compression ratios.

2025

1. NeurIPS

   HAODiff: Human-Aware One-Step Diffusion via Dual-Prompt Guidance

   Jue Gong^*, Tingyu Yang^*, Jingkai Wang, and 5 more authors

   In Advances in Neural Information Processing Systems, 2025

   Abs arXiv Website

   HAODiff addresses image restoration for human subjects under complex degradation conditions. By integrating human-aware priors with conditional guidance mechanisms, it enhances the recovery of motion-blurred body regions, maintaining perceptual quality and structural fidelity in challenging real-world degradation scenarios.