This dataset contains preprocessed full-disk Hα images (input), corresponding ground truth He I 10830 Å images, and synthetic He I 10830 Å outputs generated by a pix2pixHD deep learning model. The data are organized as follows: train_A/ – Hα training images (model inputs) train_B/ – He I 10830 Å training images (ground truth) test_A/ – Hα test images (model inputs) test_B/ – He I 10830 Å test images (ground truth) These data were used in the experiments described in: Marena, M., Li, Q., Wang, H., & Shen, B. (2025). Reconstructing He I 10830 Å Images Using Hα Images through Deep Learning. The Astrophysical Journal, 984(2), 99. https://doi.org/10.3847/1538-4357/adc7fc The dataset enables researchers to reproduce the model training and evaluation in the paper, as well as to explore alternative deep learning approaches for historical solar image reconstruction. Ackownledgement This research is supported by NASA grants 80NSSC24K0548 and 80NSSC24M0174. Hα data were acquired by GONG instruments operated by NISP/NSO/AURA/NSF with contributions from NOAA. He I 10830 Å data were acquired by SOLIS instruments operated by NISP/ NSO/AURA/NSF. Additional Hα data were provided by the Kanzelhöhe Observatory, University of Graz, Austria. We gratefully acknowledge the use of full-disk Hα data from the Big Bear Solar Observatory, supported by US NSF AGS- 2309939, and He I 10830 Å of the Mauna Loa Solar Observatory, operated by the High Altitude Observatory, as part of the National Center for Atmospheric Research (NCAR). NCAR is supported by the National Science Foundation.