LENVIZ: A High-Resolution Low-Exposure Night Vision Benchmark Dataset

TL;DR

LENVIZ introduces a large-scale, real-world low-light benchmark with multi-exposure and long-exposure frames across three camera modules, paired with expert-edited ground truth and a dedicated test set. The dataset enables both single- and multi-exposure enhancement research under diverse indoor/outdoor conditions and human content, while addressing limitations of prior datasets in size, camera diversity, and ground-truth quality. Comprehensive quantitative and qualitative evaluations show that models trained on LENVIZ achieve superior perceptual quality (LPIPS/SSIM) and robust cross-dataset generalization, complemented by a human study emphasizing brightness and naturalness. The work provides a rigorous, practical resource to advance low-light image enhancement toward production-ready, cross-camera applicability.

Abstract

Low-light image enhancement is crucial for a myriad of applications, from night vision and surveillance, to autonomous driving. However, due to the inherent limitations that come in hand with capturing images in low-illumination environments, the task of enhancing such scenes still presents a formidable challenge. To advance research in this field, we introduce our Low Exposure Night Vision (LENVIZ) Dataset, a comprehensive multi-exposure benchmark dataset for low-light image enhancement comprising of over 230K frames showcasing 24K real-world indoor and outdoor, with-and without human, scenes. Captured using 3 different camera sensors, LENVIZ offers a wide range of lighting conditions, noise levels, and scene complexities, making it the largest publicly available up-to 4K resolution benchmark in the field. LENVIZ includes high quality human-generated ground truth, for which each multi-exposure low-light scene has been meticulously curated and edited by expert photographers to ensure optimal image quality. Furthermore, we also conduct a comprehensive analysis of current state-of-the-art low-light image enhancement techniques on our dataset and highlight potential areas of improvement.

Figures (13)

• Figure 1: Long-exposure Illuminance time experimental setup. During the captures the room lighting was controlled and all the background lights were turned off to maintain the darkroom (low-light conditions).
• Figure 2: Detailed camera capture + post capture pipeline with traditional as well as LENVIZ dataset
• Figure 2: Examples of images captured during the empirical tuning of long-exposure time. Left image is DSLR captured as the control at lux meter reading=2 (post in note in scene indicates the lux reading of scene), middle image represents the long-exposure capture at exposure setting of 4 seconds, right image represents the final tuned long-exposure captured at time based on Eq \ref{['equation 2']}.
• Figure 3: Skin-tone variations and hyper-realistic mannequins.
• Figure 3: Illuminance vs exposure time (log scale) [Top] For standard values of ISO 100, aperture 1.8 and no additional boost. [Bottom] For long-exposure shot.