inaf-oact-ai/solar-flare-hmivideo-datasplits

本数据集用于训练和测试太阳耀斑预测模型，包含由Boucheron等人编译的SDO/HMI磁图图像的不同数据分割（以json格式存储），并以16帧视频形式组织，具有不同的帧速率（32分钟、76分钟）。数据集包括原始数据分割（使用32分钟、72分钟帧速率）和通过降采样获得的更平衡的数据分割，以减轻类别不平衡问题。

This dataset is intended for training and testing solar flare forecasting models. It contains various data splits (in json format) of SDO/HMI magnetogram images compiled by Boucheron, L.E., et al., 2023, Sci Data 10, 825, https://doi.org/10.1038/s41597-023-02628-8, and arranged in 16 video frames with different frame cadences (32 min, 76 min). The splits train_72min (train_36min), val_72min (val_36min), test_72min (test_36min) correspond to the original data splits provided by Boucheron et al. using different frame cadences (32 min, 72 min), while the other splits (labelled as downsampled) are created by downsampling the No-Flare and C flare class to obtain more balanced splits and mitigate class imbalance.