Tails: Chasing Comets with the Zwicky Transient Facility and Deep Learning. (arXiv:2102.13352v1 [astro-ph.IM])
<a href="http://arxiv.org/find/astro-ph/1/au:+Duev_D/0/1/0/all/0/1">Dmitry A. Duev</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Bolin_B/0/1/0/all/0/1">Bryce T. Bolin</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Graham_M/0/1/0/all/0/1">Matthew J. Graham</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Kelley_M/0/1/0/all/0/1">Michael S. P. Kelley</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Mahabal_A/0/1/0/all/0/1">Ashish Mahabal</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Bellm_E/0/1/0/all/0/1">Eric C. Bellm</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Coughlin_M/0/1/0/all/0/1">Michael W. Coughlin</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Dekany_R/0/1/0/all/0/1">Richard Dekany</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Helou_G/0/1/0/all/0/1">George Helou</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Kulkarni_S/0/1/0/all/0/1">Shrinivas R. Kulkarni</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Masci_F/0/1/0/all/0/1">Frank J. Masci</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Prince_T/0/1/0/all/0/1">Thomas A. Prince</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Riddle_R/0/1/0/all/0/1">Reed Riddle</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Soumagnac_M/0/1/0/all/0/1">Maayane T. Soumagnac</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Walt_S/0/1/0/all/0/1">Stéfan J. van der Walt</a>
We present Tails, an open-source deep-learning framework for the
identification and localization of comets in the image data of the Zwicky
Transient Facility (ZTF), a robotic optical time-domain survey currently in
operation at the Palomar Observatory in California, USA. Tails employs a custom
EfficientDet-based architecture and is capable of finding comets in single
images in near real time, rather than requiring multiple epochs as with
traditional methods. The system achieves state-of-the-art performance with 99%
recall, 0.01% false positive rate, and 1-2 pixel root mean square error in the
predicted position. We report the initial results of the Tails efficiency
evaluation in a production setting on the data of the ZTF Twilight survey,
including the first AI-assisted discovery of a comet (C/2020 T2) and the
recovery of a comet (P/2016 J3 = P/2021 A3).
We present Tails, an open-source deep-learning framework for the
identification and localization of comets in the image data of the Zwicky
Transient Facility (ZTF), a robotic optical time-domain survey currently in
operation at the Palomar Observatory in California, USA. Tails employs a custom
EfficientDet-based architecture and is capable of finding comets in single
images in near real time, rather than requiring multiple epochs as with
traditional methods. The system achieves state-of-the-art performance with 99%
recall, 0.01% false positive rate, and 1-2 pixel root mean square error in the
predicted position. We report the initial results of the Tails efficiency
evaluation in a production setting on the data of the ZTF Twilight survey,
including the first AI-assisted discovery of a comet (C/2020 T2) and the
recovery of a comet (P/2016 J3 = P/2021 A3).
http://arxiv.org/icons/sfx.gif
