LUCID: Unified Nighttime Image Restoration

• The paper introduces LUCID, a unified diffusion-based framework that addresses flare removal, noise suppression, and controlled exposure adjustment in nighttime photography.
• It employs a dual-path U-Net and conditional latent diffusion with a four-mode training regime to enable continuous HDR synthesis and fine-grained user control.
• Experimental results demonstrate that LUCID outperforms state-of-the-art methods in key metrics like CLIPIQA and MUSIQ, producing visually superior nighttime images.

LUCID, in the context of "Learning Unified Control for Image Deflaring and Exposure Mastery in Nighttime Photography," refers to a diffusion-based, modular image restoration framework that addresses the entangled degradations—specifically, strong flare and photon-limited noise—in nighttime photographic imagery. Unlike legacy approaches that attempt low-light enhancement, flare removal, or high dynamic range (HDR) synthesis as independent tasks, LUCID formulates nighttime restoration as a single, controllable process, enabling fine-grained manipulation of exposure, flare, noise, and explicit light-source appearance through classifier-free guidance. The system combines a dual-path flare disentanglement U-Net, a conditional latent diffusion pipeline, and a four-mode compositional training regime to achieve state-of-the-art performance and continuous user-driven output control (Yang et al., 5 Jun 2026).

1. Imaging Model and Problem Statement

LUCID models the observed low-light, flare-corrupted image as: $I_\mathrm{in} = R \cdot L + F + \epsilon,$ where:

• $I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$ is the observed RGB image,
• $R$ is scene reflectance, $L$ is ambient illumination (Retinex-style, multiplicative),
• $F$ is stray-light (additive lens flare and ghosting),
• $\epsilon$ is photon-limited sensor noise.

The restoration objective is to recover a noise- and flare-free, well-exposed image $I_\mathrm{tgt} \approx R \cdot L$, suppressing $F$ and removing $\epsilon$. Crucially, simply boosting exposure amplifies $F$ and $I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$0, while overly aggressive suppression of $I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$1 destroys true scene highlights.

2. Flare Disentanglement Module

LUCID’s front-end is a dual-branch U-Net that encodes $I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$2 and decodes into spatially aligned flare ($I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$3) and background ($I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$4) images: $I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$5 Decoder weights are shared for $I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$6 layers before channel-wise branching at a split layer $I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$7, producing divergent feature maps $I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$8 and $I_\mathrm{in} \in \mathbb{R}^{H \times W \times 3}$9. The loss enforces:

• Orthogonality: $R$0
• Reconstruction consistency: $R$1
• Component-wise targets: $R$2

Total disentanglement loss: $R$3 This disentanglement provides reliable structure guidance for downstream restoration.

3. Diffusion-Driven Restoration and Continuous Control

After decomposition, LUCID processes $R$4 and $R$5 using a conditional latent diffusion model:

• Images are encoded via a variational autoencoder: $R$6, $R$7.
• Two latent states are concatenated: $R$8.
• Cross-state attention is applied in each diffusion step.

Diffusion loss combines denoising, intrinsic feature ($R$9; layer features of the encoder), and LPIPS perceptual similarity: $L$0

Continuous exposure control is realized via classifier-free guidance (CFG): for scale $L$1,

$L$2

where $L$3 and $L$4 are predicted denoised latents under positive (user prompt, exposure up) and negative (omitted prompt, exposure down) conditionings. DECODING $L$5 produces the restored image, interpolating along an exposure manifold.

4. Four-Mode Training for Explicit Controllability

LUCID trains under four distinct modes, each defined by its input/conditioning tuple and control prompt:

• Mode A: Positive exposure, no light-source prompt; $L$6, prompt: “nighttime enhancement”
• Mode B: Negative exposure, no light-source prompt; $L$7, prompt: “nighttime suppression”
• Mode C: Positive exposure + light-source prompt; $L$8, prompt: “light source”
• Mode D: Negative exposure + light-source prompt; $L$9, prompt: “light source”

Minibatches randomly select among these, and classifier-free guidance is implemented by dropping prompts probabilistically (0.2 for negative conditioning, 0.5 for light-source). This setup grants explicit, disentangled control over both exposure and appearance of direct light sources.

5. Continuous HDR Reconstruction and Exposure Bracketing

LUCID supports single-input HDR by sampling the CFG exposure scale $F$0 continuously: $F$1 for $F$2. The resulting set $F$3 is fused into a composite HDR image: $F$4 where $F$5 measures local contrast and fidelity (Laplacian-pyramid fusion). This yields high-quality highlight recovery and artifact-free shadow detail from a single input.

6. Experimental Evaluation and Results

LUCID outperforms previous state-of-the-art (SOTA) methods for low-light enhancement, flare removal, and single-image HDR:

Metric	LUCID (β=1.05)	Best competitor (DarkIR)
CLIPIQA	0.4774	0.4107
MANIQA	0.3264	0.3078
MUSIQ	61.45	52.26
LIQE	3.019	2.267
NIMA	5.390	5.227

On Flare7K, LUCID eliminates streaking and over-subtraction artifacts, yielding backgrounds with realistic light-source fall-off, and sometimes improving over ground-truth. On single-image HDR, LUCID’s exposure-bracketed fusion achieves superior highlight/shadow fidelity without hallucinated structures or color artifacts. The control knob $F$6 tracks exposure value $F$7EV nearly linearly, so β in [0.25, 1.5] covers a roughly -2 to +2 EV range under standard gamma.

7. Technical Innovations and Impact

LUCID is the first unified nighttime restoration system with the following properties:

• Explicit flare-background disentanglement via a dual-decoder U-Net with orthogonality and component-wise supervision.
• Latent diffusion restoration, leveraging generative priors within a two-state, globally attentive architecture, allowing joint denoising and de-flaring.
• True continuous user control over exposure and preservation or suppression of direct light sources, realized through four-mode training and classifier-free guidance.
• Generalization to robust single-image HDR synthesis.
• Experimental superiority both in objective (no-reference IQA) metrics and visual quality, verified across diverse synthetic and real nighttime scenes.

These capabilities distinguish LUCID from conventional cascaded or fixed-parameter pipelines, directly addressing the highly entangled nature of nighttime degradations (Yang et al., 5 Jun 2026).