MANTRA and Deep-TAO Datasets

MANTRA (MAchiNe Learning Reference Lightcurve Dataset for Astronomical TRAnsients) is an annotated dataset containing 4,869 transient and 71,207 non-transient object lightcurves, built from the Catalina Real-Time Transient Survey. We provide public access to this dataset on GitHub as a plain text file to facilitate standardized quantitative comparisons of astronomical transient event recognition algorithms.

Some of the classes included in the dataset are supernovae (SN), cataclysmic variables (CV), active galactic nuclei (AGN), high proper motion stars (HPM), blazars (BZ), and flares. MANTRA can be used to experiment with various data pre-processing methods, feature selection techniques, and popular machine learning algorithms such as Support Vector Machines (SVMs), Random Forests, and Neural Networks.

The figure shows the cumulative number of lightcurves (expressed as a fraction) as a function of average magnitude (left) and the number of data points in the lightcurve (right). This includes information for the three most representative classes (SN, CV, AGN) as well as the entire dataset (ALL).

Deep-TAO (Deep-learning Transient Astronomical Object) is an annotated dataset containing 1,249,079 images, including 3,807 transient and 12,500 non-transient sequences, also built from the Catalina Real-Time Transient Survey. We created Deep-TAO to provide a clean, open, and easy-to-use dataset for benchmarking deep learning architectures in transient classification.

The transient classes included in Deep-TAO are blazars (BZ), active galactic nuclei (AGN), cataclysmic variables (CV), supernovae (SN), and other events of unknown nature. We provide public access to this dataset as FITS files in two separate GitHub repositories:

📂 Transient objects
📂 Non-transient objects

The figure shows sample images from the dataset. Each row corresponds to a sample of a different class. The temporal spacing between consecutive images varies for each example. Images were normalized for visualization.

For more information, refer to the MANTRA paper or the TAO paper.