Light field scenes with disparity maps (LFQD)

The dataset includes distorted versions of synthetic scenes created by the authors of the 4D Light Field Benchmark, as well as scenes created by us. The scenes were distorted by: Compression-based distortions (JPEG, JPEG2000, BPG, VP9, AV1, AVC, HEVC, JPEG Pleno 4D-Transform Mode) Non-compression-based distortions (Gaussian noise, impulse noise, contrast enhancement via unsharp mask) The disparity maps were estimated using: EPINET LFattNet LFDE-Flex SPO The dataset is composed of three archives. <strong>GT_scenes</strong>.zip is comprised of ground truth light field images directly created by us (the rest of the ground truths can be downloaded from https://lightfield-analysis.uni-konstanz.de). <strong>ALL_scenes</strong>.zip includes both the ground truths for scenes created by us, as well as all the distorted versions (13 scenes in total). The last archive (<strong>ALL_disparity</strong>.zip) is comprised of distorted disparity maps generated by the above-mentioned disparity estimation algorithms. The entire dataset, together with detailed information about how the scenes were distorted can be access at https://sites.google.com/fel.cvut.cz/lfqd. The dataset was created primarily for an accompanying article <em>Regarding the quality of disparity estimation from distorted light fields</em> - https://doi.org/10.1016/j.image.2022.116867.本数据集包含两类场景的失真版本：一类为4D光场基准测试（4D Light Field Benchmark）的作者所创作的合成场景，另一类为我们自主创作的场景。上述场景通过以下方式进行失真处理：基于压缩的失真（包括JPEG、JPEG2000、BPG、VP9、AV1、AVC、HEVC、JPEG Pleno 4D变换模式）、非基于压缩的失真（包括高斯噪声、脉冲噪声、通过非锐化掩码实现的对比度增强）。 本数据集的视差图采用以下四种算法进行估计：EPINET、LFattNet、LFDE-Flex、SPO。 本数据集由三个压缩包构成： <strong>GT_scenes.zip</strong> 包含我们自主创作的真值（Ground Truth）光场图像（其余真值可从https://lightfield-analysis.uni-konstanz.de下载获取）。 <strong>ALL_scenes.zip</strong> 同时收录了我们自主创作场景的真值，以及所有失真版本（共计13个场景）。 <strong>ALL_disparity.zip</strong> 包含由上述视差估计算法生成的失真相位图。 完整数据集以及关于场景失真方式的详细说明，可通过https://sites.google.com/fel.cvut.cz/lfqd 获取。 本数据集主要为配合论文<em>Regarding the quality of disparity estimation from distorted light fields</em>（DOI: 10.1016/j.image.2022.116867）而构建。