I need to develop an Machine Learning (ML) model for predicting constellation seen in a picture taken from a mobile. This I need to develop from scratch. Firstly, I need help finding a database which can provide me images of stars with constellations labelled in it. Those can be from a ground based telescope or from a camera. This is going to be a multi-labelled problem. Even after spending hours of searching I can't find a good dataset. I want a dataset with several images (preferably a few 100), each labelled with the constellation seen in them. Then I am planning to refine them and use TensorFlow for prediction. I am a grade 12 student and still new to ML, so I really need your help.


1 Answer 1


If actual photographs are not required, Stellarium could render a suitable set of reference images. Use the Perspective (gnomonic) projection for the closest match to a photograph, and display only the features you want. Its scripting capability may help you avoid repeating some manual operations.

Here are a sample pair of Stellarium images. The first has only synthetic stars, which the online plate-solving service Astrometry.net successfully recognizes:

18h+0, stars only

The second is the same part of the sky with only constellation labels and boundaries:

18h+0, constellations

The unlabeled region between Ophiuchus and Scutum is Serpens Cauda, part of the only discontinuous one of the 88 standard constellations.

  • $\begingroup$ This could be a good method but can you tell how can I render good enough resolution pictures from Stellarium especially if pictures are needed in bulk $\endgroup$ Commented Aug 16, 2020 at 5:34
  • $\begingroup$ @HarshBanka you haven't really explained what "good enough resolution" means. brightness: Do you need deep sky survey images that can see way farther and dimmer than humans and cellphones, or only images that reflect what people and hand-held cameras are likely to pick up? resolution: Do you need super high resolution images through telescopes, or only images of similar angular resolution to what people and hand-held cameras can achieve? Is it possible to clarify those things back in your original question, rather than in comments only? Thanks! $\endgroup$
    – uhoh
    Commented Aug 16, 2020 at 19:27

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .