This dataset provides a comprehensive benchmark for subjective quality assessment methodologies for the evaluation of compressed light field (LF) content in image coding scenarios. It was developed within the framework of the ISO/IEC JPEG Pleno standardization activity, specifically in support of JPEG Pleno Part 7 (Light Field Quality Assessment), and in response to the Final Call for Contributions on Subjective Light Field Quality Assessment issued by the JPEG committee (ISO/IEC JTC 1/SC29/WG1 N100307, 97th JPEG Meeting, October 2022).
The dataset comprises eight high-quality reference light fields covering a broad range of spatial and angular characteristics, including natural outdoor and indoor scenes as well as synthetic content with varying scene geometry, texture complexity, and baseline configurations. It includes both densely sampled lenslet-based light fields and sparsely sampled wide-baseline camera-array light fields, enabling the analysis of diverse spatial–angular distortions induced by compression.
The reference light fields were encoded using multiple coding solutions under carefully selected bitrate configurations to span a wide perceptual quality range. In total, 125 compressed LF stimuli were generated in accordance with the JPEG Pleno Common Test Conditions for Light Field Quality Assessment.
Subjective quality scores were obtained through controlled laboratory experiments conducted across four test labs using three subjective assessment protocols: Double Stimulus Impairment Scale (DSIS), Double Stimulus Comparison Scale with Explicit Source (DSCS-ES), and Pairwise Comparison (PC). Light fields were visualized using pseudo-video rendering along predefined angular trajectories (spiral and serpentine), ensuring controlled and repeatable viewing conditions. All experiments were performed on calibrated 4K displays under standardized viewing conditions.
In case you use this dataset in your research, please cite the following paper:
