Reproduce “A Perceptually Optimized and Self-Calibrated Tone Mapping Operator” 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

Peibei Cao, Chenyang Le, Yuming Fang, Kede Ma

A two-stage deep learning tone mapping operator (PS-TMO) integrates perceptual optimization with self-calibration for high dynamic range images, producing LDR outputs that ranked first in subjective quality evaluations. The PS-TMO also achieved the highest average TMQI score of 0.9509 and processed images in real-time on standard GPU hardware.

Key insights

  • Perceptual metrics like NLPD and MEF-SSIM can serve as effective loss functions to train DNNs for cross-dynamic-range image quality, surpassing limitations of synthetic ground truths.
  • A self-calibration mechanism, inspired by multi-exposure image fusion principles, enables automatic processing of uncalibrated HDR images by generating and fusing a "pseudo-multi-exposure" stack.
  • Combining lightweight deep neural networks with a physiology-driven image decomposition (normalized Laplacian pyramid) achieves high visual quality and real-time computational performance.

Summary by alphaXiv.

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