The Pytorch implementation of LIE-IQA.

You can get the MindSpore implementation here LIE-IQA-mindspore. It is worth noting that the MindSpore implementation of Image Intrinsic Decomposition (IID) Module is different from the Pytorch implementation , but there is not much difference in performance. Please refer to the specific code for details.

The code on Pytorch will be released soon.

• Python3
• Pytorch 1.6
• Torchvision
• cuda 10.2

• LIE-IQA Framework

  

• Performance on LIE-IQA Dataset

  

• Performance on General Scene IQA Dataset

  • LIVE, MDID, CSIQ
  

• Optimization framework

  

• Qualitative comparison of the quality optimization result

  • DALE[1]
  

• Quantitative comparison of the quality optimization result

  • SSIM, NIQE, DISTS, hyperIQA and our LIE-IQA
  

[1]Kwon, Dokyeong, Guisik Kim, and Junseok Kwon. "DALE: Dark Region-Aware Low-light Image Enhancement." arXiv preprint arXiv:2008.12493 (2020).