Deep Tone Mapping Operator for High Dynamic Range Images

摘要: A computationally fast tone mapping operator (TMO) that can quickly adapt to a wide spectrum of high dynamic range (HDR) content is quintessential for visualization on varied low (LDR) output devices such as movie screens or standard displays. Existing TMOs successfully tone-map only limited number HDR and require an extensive parameter tuning yield the best subjective-quality tone-mapped output. In this paper, we address problem by proposing fast, parameter-free scene-adaptable deep (DeepTMO) yields high-resolution high-subjective quality mapped Based conditional generative adversarial network (cGAN), DeepTMO not learns vast scenic-content (e.g., outdoor, indoor, human, structures, etc.) but also tackles related scene-specific challenges contrast brightness, while preserving fine-grained details. We explore 4 possible combinations Generator-Discriminator architectural designs specifically some prominent issues in deep-learning frameworks like blurring, tiling patterns saturation artifacts. By exploring different influences scales, loss-functions normalization layers under cGAN setting, conclude with adopting multi-scale model our task. To further leverage large-scale availability unlabeled data, train generating targets using objective metric, namely Tone Mapping Image Quality Index (TMQI). demonstrate results both quantitatively qualitatively, showcase generates high-resolution, high-quality images over large real-world scenes. Finally, evaluate perceived conducting pair-wise subjective study which confirms versatility method.