Abstract:Glare is a significant concern when implementing passive daylighting strategies because it often impedes their effectiveness. The high dynamic range (HDR) method, which is employed to generate luminance maps for glare detection, aids in the real-time control of shading devices for glare mitigation. However, the HDR method entails considerable processing time and complexity, which limits its practical utility. Consequently, we propose a two-step network that combines generative adversarial networks (GANs) and reconstruction-net to transform a single low dynamic range (LDR) image into a comprehensive fisheye luminance map. This methodology offers a pragmatic approach that significantly streamlines the luminance map generation workflow and reduces the time needed for image acquisition. Additionally, we introduce a GAN generator model to recover missing pixel values in underexposed and overexposed LDR images. To evaluate our proposed network, we assembled a fisheye luminance map dataset that comprises 884 scenes, each containing 15 exposure values ranging from −7 to +7. We compared our method with existing models by using the test dataset and obtained a peak signal-to-noise ratio (PSNR) and R 2 of daylight glare probability (DGP) of 59.24 and 0.9054, respectively. These results demonstrate that our proposed network represents a state-of-the-art (SOTA) model in luminance map restoration. Furthermore, our model processes each input image in approximately 0.1 s on an RTX 2060 GPU and 7.5 s on an i5-9400F CPU, with an additional 2 s needed by Radiance to calculate the glare metric on an i5-9400F CPU. Our model demonstrates a remarkable 12-fold decrease in processing time compared to traditional methods when run on a CPU. This efficiency dramatically increases, up to a 95-fold reduction, when a GPU is utilized for reconstruction. Our code and dataset are available at: https://github.com/ShikangWen2000/SingleLM-Net.git .

Reconstructing fisheye luminance maps with a two-step network from a single low dynamic range image

From Coarse to Fine: A Two-Stage Network for Dense Haze removal

Luminance Attentive Networks for HDR Image and Panorama Reconstruction

Color and Luminance Separated Enhancement for Low-Light Images with Brightness Guidance

Deep Neural Network Approach for Annual Luminance Simulations

Adaptive Light Estimation using Dynamic Filtering for Diverse Lighting Conditions

Nighttime Visibility Enhancement by Increasing the Dynamic Range and Suppression of Light Effects

A Light Dual-Task Neural Network for Haze Removal

Sparse Gradient Regularized Deep Retinex Network for Robust Low-Light Image Enhancement

Rendering the Image of Glare Effect Based on Paired and Unpaired Dual Generative Adversarial Network

Generation of High Dynamic Range Illumination from a Single Image for the Enhancement of Undesirably Illuminated Images

RDGAN: Retinex Decomposition Based Adversarial Learning for Low-Light Enhancement

A High-Performance Face Illumination Processing Method via Multi-Stage Feature Maps

Deep Retinex Network for Single Image Dehazing

HDR Light Field Imaging of Dynamic Scenes: A Learning-based Method and A Benchmark Dataset

Remote sensing image dehazing using generative adversarial network with texture and color space enhancement

Multi-path parallel enhancement of low-light images based on multiscale spatially aware Retinex decomposition

Hybrid Saturation Restoration for LDR Images of HDR Scenes

LAN-HDR: Luminance-based Alignment Network for High Dynamic Range Video Reconstruction

Progressive and Selective Fusion Network for High Dynamic Range Imaging