Learning a Practical SDR-to-HDRTV Up-conversion using New Dataset and Degradation Models

In media industry, the demand of SDR-to-HDRTV up-conversion arises when userspossess HDR-WCG (high dynamic range-wide color gamut) TVs while mostoff-the-shelf footage is still in SDR (standard dynamic range). The researchcommunity has started tackling this low-level vision task by learning-basedapproaches. When applied to real SDR, yet, current methods tend to produce dimand desaturated result, making nearly no improvement on viewing experience.Different from other network-oriented methods, we attribute such deficiency totraining set (HDR-SDR pair). Consequently, we propose new HDRTV dataset (dubbedHDRTV4K) and new HDR-to-SDR degradation models. Then, it's used to train aluminance-segmented network (LSN) consisting of a global mapping trunk, and twoTransformer branches on bright and dark luminance range. We also updateassessment criteria by tailored metrics and subjective experiment. Finally,ablation studies are conducted to prove the effectiveness. Our work isavailable at: https://github.com/AndreGuo/HDRTVDM.