Compare Tikhonet to SCORE

[fadinammour]

Goal : establish a comparison between Tikhonet vs Tikhonet+Shape Constraint vs Sparse Deconvolution vs Sparse Deconvolution+Shape Constraint (SCORE)

• extract a data set for the test.
• run tests on each method.
• set metrics.
• analyse results.

[fadinammour]

As a first approach, we can start by extracting 2048 images (by ideally setting seeds for Numpy and TensorFlow random generators) and saving them. However, on the long run, it would be better to have more galaxies.
By doing so, we would be able to compare different approaches on the exact same objects.

[fadinammour]

For the time being the dataset that is generated has been filtered by removing the galaxies which window estimation failed. These galaxies usually correspond to unresolved galaxies which shape study is not relevant (for more details see issue 12). The filtering processing removes 33-40% of the galaxies.
In this notebook, an initial batch of 128*24 = 3072 is generated. 40% of the galaxies have a window failure which give a finale size of roughly 1840 galaxy.