Reproduce “Explorable Tone Mapping Operators” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chien-Chuan Su, Ren Wang, Hung-Jin Lin, Yu-Lun Liu, Chia-Ping Chen, Yu-Lin Chang, Soo-Chang Pei
A deep learning approach generates diverse tone-mapped variations from a single HDR image, enabling users to explore different visual styles. The method produces visually appealing results with minimal artifacts and achieves competitive performance against state-of-the-art methods on metrics like TMQI.
Key insights
- Leveraging multimodal image-to-image translation with the BicycleGAN framework enables the generation of diverse tone mapping styles from a single input, addressing the inherent subjectivity of optimal tone mapping.
- Decomposing HDR images into learned base and detail layers using a Kernel Prediction Network helps preserve fine details and avoid common artifacts more effectively than traditional handcrafted filters.
- Conditioning both the decomposition and tone compression networks on random latent codes allows for continuous exploration of a wide range of tone mapping styles in a controllable manner.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.