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

2026-06-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, extended

Summary

LUCID is a novel unified framework designed for nighttime photography restoration, addressing the entangled degradations of intense lens flares and photon-limited noise. It reframes restoration as a continuous, controllable process, rather than a fixed correction. The framework comprises a Flare Disentanglement Module, which isolates optical artifacts, and a Diffusion-Driven Restoration Module, leveraging generative priors for clean, well-exposed imagery. LUCID introduces a four-mode training strategy, enabling explicit controllability over light sources, flare, ghosting artifacts, and continuous exposure via classifier-free guidance (CFG). This allows for single-image High Dynamic Range (HDR) reconstruction. Extensive experiments demonstrate LUCID's superior performance over state-of-the-art methods across diverse real-world nighttime scenarios, including general enhancement, flare mitigation, and HDR.

Key takeaway

For computer vision engineers developing image processing pipelines or photographers seeking advanced post-production tools, LUCID offers a robust solution. Its unified, controllable approach to nighttime image restoration, including continuous exposure and selective flare management, means you can achieve superior results and creative flexibility that traditional cascaded methods or generic AI tools cannot match. Consider integrating LUCID to overcome the limitations of entangled degradations and gain precise artistic control.

Key insights

LUCID unifies controllable flare mitigation and exposure adjustment for superior nighttime image restoration.

Principles

• Nighttime degradations are fundamentally entangled, requiring cooperative resolution.
• Decompose additive flare from underlying scene content for targeted suppression and structural guidance.

Method

LUCID uses a U-Net for flare disentanglement, then a mixing-state diffusion U-Net with a four-mode training strategy and Classifier-Free Guidance (CFG) for continuous exposure and light-source control.

In practice

• Adjust image brightness and light source appearance continuously using CFG.
• Synthesize pseudo-exposure sequences from a single image for HDR reconstruction.
• Use as a pre-processor for generative AI to remove artifacts or a post-processor for exposure tuning.

Topics

• Nighttime Photography
• Lens Flare Removal
• Diffusion Models
• Image Restoration
• High Dynamic Range
• Controllable AI

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• LUCID: Learning Unified Control for Image Deflaring and Exposure Mastery in Nighttime Photography — LUCID unifies nighttime image restoration by controlling entangled flares and noise through a continuous process.
• IG-Diff: Complex Night Scene Restoration with Illumination-Guided Diffusion Model — Illumination-guided diffusion models effectively restore complex night scenes by integrating illumination priors and synthetic data.
• DuET: Dual Expert Trajectories for Diffusion Image Editing — DuET improves diffusion image editing by temporarily relaxing source conditioning, enhancing fidelity with a minor preservation trade-off.
• PIXLRelight: Controllable Relighting via Intrinsic Conditioning — PIXLRelight enables fast, high-quality single-image relighting by integrating PBR with learned image synthesis via intrinsic conditioning.
• How this AI Learned to Refocus Any Photo — AI refocusing allows dynamic post-capture adjustment of photographic focus and depth of field from a single image.
• NVIDIA’s Insane AI Found The Math Of Reality — NVIDIA's PPISP technique improves NeRFs by mathematically correcting camera-induced distortions in input images.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.