Bidirectional Translation Between UHD-HDR and HD-SDR Videos

With the popularization of ultra high definition (UHD) high dynamic range (HDR) displays, recent works focus on upgrading high definition (HD) standard dynamic range (SDR) videos to UHD-HDR versions, aiming to provides richer details and higher contrasts on advanced modern displays. However, joint considering the upgrading & downgrading translations between two types of videos, which is practical in real applications, is generally neglected. On the one hand, downgrading translation is the key to showing UHD-HDR videos on HD-SDR displays. On the other hand, considering both translations enables joint optimization and results in high quality translation. To this end, we propose the bidirectional translation network (BiT-Net), which jointly considers two translations in one network for the first time. In brief, BiT-Net is elaborately designed in an invertible fashion that can be efficiently inferred along forward and backward directions for downgrading and upgrading tasks, respectively. Based on this framework, we divide each direction into three sub-tasks, i.e., decomposition, structure-guided translation, and synthesis, to effectively translate the dynamic range and the high-frequency details. Benefiting from the dedicated architecture, our BiT-Net can work on 1) downgrading UHD-HDR videos, 2) upgrading existing HD-SDR videos, and 3) synthesizing UHD-HDR versions from the downgraded HD-SDR videos. Experiments show that the proposed method achieves state-of-the-art performances on all these three tasks.

I. Introduction

Recently, new video standards [1], [2], [3] (e.g., HDR10, Dolby Vision) have been proposed to specify ultra high definition (UHD) and high dynamic range (HDR) as prerequisites of compliance, which provide better visual experience on modern HDR displays. Compared to ordinary high definition (HD) videos of standard dynamic range (SDR), UHD-HDR videos remarkably enhance the video appearance from various aspects such as larger spatial resolution (3840 × 2160), larger color depth (10 bits), and new color space (BT.2020).