Overview The DeepSDO description dataset is a collection of pairs consisting of SDO fulldisk solar images and corresponding captions. This dataset covers nine solar features: sunspots, flares, prominences, prominence eruptions, coronal holes, coronal loops, filaments, active regions, and eclipses. The experts in solar physics fields referred to solar events observed from 2010 to 2019 using SDO AIA and HMI, and the SDO gallery, provided by NASA, served as a reference for selecting and publishing detailed descriptions and images of noteworthy or particularly large solar events. We provides pairs of SDO fulldisk solar JPG images and ground-truth captions generated by humans to train and evalutate machine learning models in understanding and analyzing solar phenomena.

Image Captioning

9 solar events (sunspots, flreas, prominences, prominence eruptions, coronal holes, coronal loops, filaments, active regions, and eclipses)

1051 pairs of images and captions

Download Train and Test Dataset: open soon

Research Paper Deep Learning-based Solar Image Captioning (Baek et al. 2024)

Terms of Use The annotations and images in this dataset belong to the Korean Data Center for SDO of KASI.

Citing Dataset in Publications The dataset citation should be to the research paper (Baek et al. 2024)