The 35th annual conference on Astronomical Data Analysis Software & Systems 2025 in Görlitz, Germany

We present the Bayesian framework "Incliscope", a novel approach to estimate inclinations of galaxies based on optical images. In contrast to traditional methods, our solution does not rely on the fitting of ellipsoids in order to solve the axis-ratio equation. Instead, we use a probabilistic approach to collect properly-calibrated posterior distributions among inclinations from simulated galaxies and use them as training data for a Deep Convolutional Mixture Density Network.

The distributions predicted by the trained network reach an average accuracy of 4.3° compared to the respective true inclination angles. Our method outperforms the isophote-fitting technique in measures of accurary, stability and time-efficiency.