Dual-Input Neural Networks for Personalized Image Precompensation

This dataset supports the paper "Dual-Input Neural Networks for Personalized Image Precompensation" (J. Imaging, 2025).
It includes color images and point spread functions (PSFs) used to train and evaluate deep neural networks that compensate for human visual distortions caused by refractive errors.

📂 Dataset structure:

• train/

  • 1000 color images from ImageNet-21k (resized to 512×512)

  • PSFs from three categories: narrow, medium, and broad

• test/

  • 735 color test images from the SCA-2023 dataset (6 categories)

  • PSFs from three categories: narrow, medium, and broad  (unseen during training)

• PSFs are generated from ophthalmic parameters: spherical (S), cylindrical (C), and axis (A)

This dataset enables reproducible training and benchmarking of dual-input neural networks for personalized image precompensation, with full control over PSF diversity and severity.