jorgemarpa/tess-asteroid-ml: v0.2.0

TESS Asteroid cutouts for ML A library to create ML-ready data from TESScut with asteroids This Python package provides an API and scripts to create cutouts from TESS full frame images and query JPL Horizon to find observed asteroids in the FFI images.The resulting data are the flux cube, asteroid mask in the same shape as the flux cube, and other related vectors (time, CBV, quaternions, etc) that are formated to be used by ML detection models. Installation Clone this repo and then install it with pip clone https://github.com/jorgemarpa/tess-asteroid-ml.git cd tess-asteroid-mlpip install . Example use To create the ML ready training data use the *make_TESScut_asteroids.py* script with the desired arguments (see below). This will do the following: 1. First access the asked TESS Sector/Camera/CCD image from MAST/AWS. 2. Given a grid pattern (sparse or dense) it uses AstroCut to make cutouts from the FFI and save them to disk. Be carefull to run the script with the argument `--download` only the first time for a Sector/Camera/CCD because it will query Astrocut to make and download the files. Fowllowing runs must be withouth the flag to use local data.3. Uses the asteroid catalog to get the first guess of available asteroids in the FFI.4. Iterates over the available cutouts and available asteroids to make the mask arrays and access other vectors (CBV, quaternions, time, etc). 5. It saves the training data as *npz* files. When runing the script the first time for a Sector/Camera/CCD use: python make_TESScut_asteroids.py --sector 4 --camera 3 --ccd 2 --sampling sparse --cutout-size 64 --lim-mag 22 --verbose --plot --download The script has several arguments and flags. Below is the full list: usage: make_TESScut_asteroids.py [-h] [--sector SECTOR] [--camera CAMERA] [--ccd CCD] [--cutout-size CUTOUT_SIZE] [--lim-mag LIM_MAG] [--sampling SAMPLING] [--fit-bkg] [--plot] [--verbose] [--download] Creates a dataset from TESS FFIs. Makes 64x64 pixel cuts, uses JPL to create a asteroid mask. optional arguments: -h, --help show this help message and exit --sector TESS sector number. --camera TESS camera number. --ccd TESS CCD number --cutout-size Cutout size in pixels --lim-mag Limiting magnitude in V band. --sampling Select a `dense` grid that covers corner to corner of the FFI or a `sparse` that uses only 7 rows from the grid. --fit-bkg Fit and substract background (flag). --plot Plot FFI (flag). --verbose Verbose (flag). --download Donwload cutouts from from AWS wh Astrocut (flag). Citation If you use this dataset or code in your research, please cite it as follows: Martínez-Palomera, J. I. (2026). TESS Asteroid cutouts for ML. Zenodo. [https://doi.org/10.5281/zenodo.19558224](https://doi.org/10.5281/zenodo.19558224) Powell et al. (AJ 2026). Trajectory-Agnostic Asteroid Detection in TESS with Deep Learning Contact For questions regarding the methodology or data, please contact the repository administrator. Keywords: Astrophysics, Machine Learning, Semantic Segmentation, TESS, Roman Space Telescope, Asteroids, W-net. Communities: NASA Goddard Space Flight Center, CRESST II, Open Astronomy.