DeepMerge II: Building Robust Deep Learning Algorithms for Merging Galaxy Identification Across Domains (Data)

We present the data used in "DeepMerge II: Building Robust Deep Learning Algorithms for Merging Galaxy Identification Across Domains". In this paper, we test domain adaptation techniques, such as Maximum Mean Discrepancy (MMD) and adversarial training with Domain Adversarial Neural Networks (DANNs) for cross-domain studies of merging galaxies. Domain adaptation is performed between two simulated datasets of various levels of observational realism (simulation-to-simulation experiments), and between simulated data and observed telescope images (simulation-to-real experiments). For more details about the datasets please see the paper mentioned above.

Simulation-to-Simulation Experiments

Data used to study distant merging galaxies using simulated images from the Illustris-1 cosmological simulation at redshift z=2. The images are 75x75 pixels with three filters applied that mimic Hubble Space Telescope (HST) observations (ACS F814W,NC F356W, WFC3 F160W) with added point-spread function (PSF) and with or without observational noise.

Source Domain

     Images: SimSim_SOURCE_X_Illustris2_pristine.npy

     Labels: SimSim_SOURCE_y_Illustris2_pristine.npy

Target Domain 

    Images: SimSim_TARGET_X_Illustris2_noisy.npy

    Labels: SimSim_TARGET_y_Illustris2_noisy.npy

Simulation-to-Real Experiments

Data used to study nearby merging galaxies using simulated Illustris-1 images at redshift z=0 and observed Sloan Digital Sky Survey (SDSS) images from the Galaxy Zoo project. All images have three filters. SDSS images have (g,r,i) filters, while simulated Illustris images also mimic the same three SDSS filters with added effects of dust, PSF and observational noise.

Source Domain 

    Images: SimReal_SOURCE_X_Illustris0.npy

    Labels: SimReal_SOURCE_y_Illustris0.npy

Target Domain 

    Images: SimReal_TARGET_X_postmergers_SDSS.npy

    Labels: SimReal_TARGET_y_postmergers_SDSS.npy