Reproduce “Image Correction via Deep Reciprocating HDR Transformation” 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
Xin Yang, Ke Xu, Yibing Song, Qiang Zhang, Xiaopeng Wei, Rynson Lau
This paper proposes a Deep Reciprocating HDR Transformation (DRHT) framework that reconstructs and corrects images by operating within the High Dynamic Range (HDR) domain, allowing the recovery of details lost in severely under- or overexposed photographs. The method achieves superior quantitative and qualitative results, including a PSNR of 28.18 on the City Scene dataset, substantially outperforming both traditional Low Dynamic Range (LDR) methods and existing HDR prediction techniques.
Key insights
- Details that appear lost in Low Dynamic Range (LDR) images due to exposure issues are often still present in the scene's High Dynamic Range (HDR) data.
- True detail recovery for severely exposed images necessitates operating beyond the limitations of the LDR domain.
- By first reconstructing and then manipulating images within the HDR domain, information inaccessible to traditional LDR-based image correction methods can be successfully restored.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.