FastHDRNet: A new efficient method for SDR-to-HDR Translation

现代显示器现在具备高动态范围（HDR）和广泛色域的视频内容渲染能力。然而，大部分可用资源仍处于标准动态范围（SDR）。因此，我们需要确定一种有效的方法。现有的基于深度神经网络（DNN）的SDR到HDR转换方法优于传统方法，但它们要么太大无法实现，要么会生成一些可怕的伪影。我们提出了一种SDR到HDR转换的神经网络，称为“FastHDRNet”。该网络包括两个部分，自适应通用颜色转换（AUCT）和局部增强（LE）。该架构被设计为轻量级网络，利用全局统计和局部信息具有超高效率。经过实验，我们发现我们提出的方法在定量比较和视觉质量方面均实现了最先进的性能，同时具有轻量级结构和增强的推理速度。

SDR to HDR conversion using neural networks

Proposing a lightweight neural network for SDR to HDR conversion, utilizing global statistics and local information with high efficiency

The proposed FastHDRNet achieves state-of-the-art performance in both quantitative comparisons and visual quality with a lightweight structure and enhanced infer speed. The network includes Adaptive Universal Color Transformation and Local Enhancement. Experimentation is done with various datasets and the results are promising. The proposed method can be used in real-time applications.

Related studies in this field include 'Learning to See in the Dark' by Chen et al., and 'Deep Bilateral Learning for Real-Time Image Enhancement' by Chen and Paris.