Reproduce “Unsupervised HDR Image and Video Tone Mapping via Contrastive Learning” 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

Cong Cao, Huanjing Yue, Xin Liu, Jingyu Yang

Researchers from Tianjin University introduced IVTMNet, an unsupervised deep learning framework for high-quality tone mapping of HDR images and videos to LDR. It established new state-of-the-art performance in image tone mapping and delivered pioneering results for video, achieving improved perceptual quality and temporal consistency without requiring paired HDR-LDR datasets.

Key insights

  • Unsupervised contrastive learning, augmented by using poorly-exposed LDRs as negative samples and leveraging an intrinsic quality metric (TMQI) for instance-level guidance, effectively trains the model without paired data.
  • Specific latent codes constructed from mean brightness and contrast of feature channels, combined with a custom similarity metric, are more effective for perceptual tone mapping tasks than generic feature representations.
  • The MAC-free Temporal-Feature-Replaced (TFR) module efficiently enhances temporal consistency in videos by selectively replacing feature portions from previous frames, mitigating flickering artifacts.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.