I was reading Deep-CEE I: Fishing for Galaxy Clusters with Deep Neural Nets and I couldn't find the dataset.

Could someone provide me with the links related to this?

  • 1
    $\begingroup$ Have you had a chance to look at my answer? $\endgroup$ Aug 11, 2020 at 22:37
  • 1
    $\begingroup$ @WilliamMiller I did! :-) $\endgroup$
    – uhoh
    Aug 12, 2020 at 19:02

1 Answer 1


The authors used the SDSS (Sloan Digital Sky Survey) III dataset (data release 9):

We use a state-of-the-art probabilistic algorithm, adapted to localise and classify galaxy clusters from other astronomical objects in SDSS imaging.

This data can be accessed through Nasa SkyView. The training and test sets were constructed from subsets of the Abell catalogue:

We use the Abell galaxy clusters identified in the Wen et. al (2012) catalogue, to obtain the labelled data needed to create the training set.

Most of the clusters in the Abell catalogue have been verified, but not all. Hence the use only of those which are included in Wen et. al (2012). The clusters were filtered based on the following criteria (section 2.2):

  • Photometric redshift ($z$) range limited to $0.1 < z < 0.2$.
  • Minimum of 20 galaxy members within $R_{200}$ radius.

$497$ Abell clusters satisfy these criteria. The authors applied translational shifts to augment the data:

... one of the properties of the FasterRCNN algorithm is translational invariance, which means the algorithm is robust at learning translated objects. We train the algorithm to recognise that an object could appear at any location in an image. Since our method applies a random offset to the input coordinates via translation we augment the sample set three additional times, which boosts the sample size to $1988$.

This set of $1988$ inputs was randomly sampled to produce the training and test sets:

The training set is made up of $∼90$ percent of the sample set consisting of $1784$ labelled galaxy clusters and the test set is made up of the remaining $∼10$ per cent consisting of $204$ labelled galaxy clusters.

The authors do not provide the translated data. However, since the process is statistically robust, applying three different set of random translations to the $497$ images and running the model should produce results which are consistent with those reported. This is, in fact, a fundamental assumption of the technique.

  • $\begingroup$ Thanks a lot :) $\endgroup$
    – Madara
    Aug 13, 2020 at 4:37

You must log in to answer this question.

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