HDR to SDR with AI
Tone mapping operators with AIHigh Dynamic Range (HDR) is fast becoming the video format of choice for content providers, albeit not quite as quickly as thought. Standard Dynamic Range (SDR) displays still dominate the market. Therefore, efficient ways must be found to convert HDR content to SDR format, a process known as tone mapping. There are already numerous tone mapping operators (TMO for short), but these do not always lead to a usable result.
Recently, many "new" TMOs based on deep learning approaches have been proposed. However, the main challenge in training such deep learning networks is the lack of true-to-life SDR and HDR datasets, which would result in highly accurate TMOs.
A recent article on Generative Adversarial Network Based Tone Mapping Operator for 4K HDR Images presented a high-quality 4K HDR-SDR dataset of image pairs covering a wide range of brightness levels and colors. A TMO based on a generative adversarial network architecture was proposed.
"The evaluation results demonstrate that our method achieves high perceptual quality, maintains artistic intent, and provides better color representation compared to existing state-of-the-art TMOs." (Quote end)
For anyone interested:
Data and code are available at https://github.com/zjbthomas/TMO-GAN