Publication:
The xDR dataset: A cinematic natively graded HDR & SDR dataset for evaluation of inverse tone mapping methods
Date
2026
Journal article
Loading...
Author(s)
Journal
SIGNAL PROCESSING-IMAGE COMMUNICATION
Abstract
High dynamic range (HDR) displays can show visual content with higher brightness and deeper black levels than traditional standard dynamic range (SDR) displays. This capability allows HDR displays to produce more vivid and realistic visual content that closely resembles real-world scenarios. Despite the advancements in HDR technology, SDR content remains prevalent in home video distribution and broadcasting, with a substantial amount of legacy content stored in the SDR format. To appropriately display SDR content on HDR displays, its dynamic range must be expanded using an inverse tone mapping operator (iTMO). A common method to evaluate the performance of iTMOs involves comparing the HDR content they produce with actual HDR reference content. This evaluation process, known as Image Quality Assessment (IQA) or Video Quality Assessment (VQA), depending on the type of content, typically requires a test dataset that includes visual content – images or video sequences – in HDR and SDR. However, a significant limitation of existing datasets is the lack of visual content created equally in both SDR and HDR following a cinematic production pipeline. This often results in artistic intentions not being equally expressed in both formats, which means that quality assessment studies based on these datasets do not accurately reflect the performance of iTMOs in professional content production environments, where visual content is carefully graded using calibrated monitors. To address this gap, this paper presents a new publicly available dataset, the “xDR Dataset”. Our dataset includes 10 video sequences, each approximately 40 s long, in full HD (1920×1080). These video sequences were captured and natively graded in HDR and SDR by professionals, ensuring that both versions express the same artistic intentions. The sequences in our dataset encompass a wide variety of cinematic scenes with different contrasts and lighting conditions, representing challenging scenarios for inverse tone-mapping algorithms. Additionally, we present a study using this dataset to assess the performance of the most recent inverse tone mapping methods in creating visual content with perceivable quality comparable to actual cinematic HDR content.