A High-Resolution High Dynamic Range Light-Field Dataset with an Application to View Synthesis and Tone-Mapping

Abstract

Light-field (LF) imaging has various advantages over the traditional 2D photography, providing angular information of the real world scene by separately recording light rays in different directions. Despite the directional light information which enables new capabilities such as depth estimation, post-capture refocusing, and 3D modelling, currently available light-field datasets are very restricted in terms of spatial-resolution and dynamic range. In this work, we address this problem by capturing a novel light-field dataset featuring both a high spatial resolution and a high dynamic range (HDR). This dataset should enable the community to research and develop efficient reconstruction and tone-mapping algorithms for a hyper-realistic visual experience. The dataset consists of six static light-fields that are captured by a high-quality digital camera mounted on two precise linear axes using exposure bracketing at each view point. To demonstrate the usefulness of such a dataset, we also performed a thorough analysis on local and global tone-mapping of natural data in the context of novel view-rendering. The rendered results are compared and evaluated both visually and quantitatively. To our knowledge, the recorded dataset is the first attempt to jointly capture high-resolution and HDR light-fields.