A CNN to classify images different types of galaxies - spiral, elliptical, and irregular.

Trained and tested on 16 gigs of RAM, i7-8750H, GTX 1060 all at stock settings.

Check requirements.txt and packages_neded.txt for module information.

• First generate the training command using generate_training_command.py. The command relies on specifying the absolute path so the script will generate the command specific to your environment.

  python generate_training_command.py

• Then execute the command which will be of this form:

  python your_project_path\retrain.py --bottleneck_dir=your_project_path\bottlenecks --how_many_training_steps 500 --model_dir=your_project_path\inception --output_graph=your_project_path\retrained_graph.pb --output_labels=your_project_path\retrained_labels.txt --image_dir your_project_path\tf_files

  You can change the number of training steps and other parameters inside the generate_training_command.py file.

python label_image.py test_images/ellipticaltest.jpg

Output:

elliptical (score = 98.54690)
spiral (score = 0.86201)
irregular (score = 0.59109)