BTSbot-maxvit-tiny-randinit

BTSbot

This is a maxvit fine-tuned for classifying images from the Zwicky Transient Facility (ZTF) observatory. Rehemtulla et al. 2024 originally introduced BTSbot and its classification task, and Rehemtulla et al. 2025 performed architecture and pre-training benchmarking on this BTSbot image classification task.

Base Model: timm/maxvit_tiny_rw_224.sw_in1k

Usage

Easily install the btsbot package and load this model with:

pip install btsbot

import btsbot
model = btsbot.load_HF_model(
    architecture="maxvit", multi_modal=False, pretrain="randinit"
)

Also see BTSbot/btsbot/inference_example.py.

Citation

If you use this model, please cite:

@ARTICLE{Rehemtulla+2025,
       author = {{Rehemtulla}, Nabeel and {Miller}, Adam A. and {Walmsley}, Mike and {Shah}, Ved G. and {Jegou du Laz}, Theophile and {Coughlin}, Michael W. and {Sasli}, Argyro and {Bloom}, Joshua and {Fremling}, Christoffer and {Graham}, Matthew J. and {Groom}, Steven L. and {Hale}, David and {Mahabal}, Ashish A. and {Perley}, Daniel A. and {Purdum}, Josiah and {Rusholme}, Ben and {Sollerman}, Jesper and {Kasliwal}, Mansi M.},
        title = "{Pre-training vision models for the classification of alerts from wide-field time-domain surveys}",
      journal = {arXiv e-prints},
     keywords = {Instrumentation and Methods for Astrophysics, Computer Vision and Pattern Recognition},
         year = 2025,
        month = dec,
          eid = {arXiv:2512.11957},
        pages = {arXiv:2512.11957},
          doi = {10.48550/arXiv.2512.11957},
archivePrefix = {arXiv},
       eprint = {2512.11957},
 primaryClass = {astro-ph.IM},
       adsurl = {https://ui.adsabs.harvard.edu/abs/2025arXiv251211957R},
      adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}

License

This model is released under the MIT License.

Repository

For more information, see the BTSbot GitHub repository.