Astrophysics Goals ================== - Star formation as a fonction of the environment (physical conditions) - Bridging the gap between Galactic & extragalactic star formation - Extragalactic SF: Use the built model + change the spatial resolution .. container:: How SF depends on the environment in galaxies ? Galactic & extragalactic star formation ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ .. image:: _static/figastro1.png :align: center :alt: star formation SF components ^^^^^^^^^^^^^ Filaments, clumps, cores, young stars 2D filament Segmentation ^^^^^^^^^^^^^^^^^^^^^^^^ .. image:: _static/segment1.png :align: center :alt: 2D segmentation Method: supervised learning *************************** - Database : Hi-GAL column density maps: 32 000 filaments + their associated RoI - Images of different size - For each filament: 3 patches (64x64) are taken from 3 random positions around the RoI - Must contain ≥ 20% of the mask - Parts with missing data in the original image are not taken (saturation zones) - Data augmentation: 2 flips and 3 rotations .. container:: ~ 900 000 patches + their associated RoI - U-Net5 Convolutional Neural Network used for image segmentation - Base: 80% training set and 20% validation set - Python + Keras: 20 epochs for the learning phase .. image:: _static/unet.png :align: center :alt: 2D segmentation