hDRIVE: HDR Image Visual Evaluation Metric for SDR to HDR Upconversion Quality Assessment

ICLR 2026 Conference Submission25592 Authors

Keywords: HDR, SDR, inverse tone mapping, video quality assessment

Abstract: HDR displays are becoming increasingly common on both TVs and mobile devices, that requires to adapt existing legacy SDR to HDR screens. Several algorithms have been developed for SDR-to-HDR upconversion, also known as Inverse Tone Mapping (ITM). However, there is still a lack of reliable metrics for assessing the quality of these algorithms. This is due in part to the ill-posed nature of the ITM task, where the most visually pleasing result can significantly differ from the original image. In this work, we propose a novel state-of-the-art no-reference video quality metric for evaluating upconverted HDR content. To support our approach, we collect a large-scale dataset of human visual preferences, capturing both the perceived visual appearance and quality of HDR videos. The HDR ITM video quality metric might be very helpful and drive the rapid advancement of SDR-to-HDR algorithms development and benchmarking. Both the metric and the training dataset are publicly available for download via the provided link.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 25592