Galaxies Image Classifier

Project description

The goal of this project is to classify galaxies images into 10 different classes. In order to achieve this goal, several models have been trained and tested, as explained in the report.

The final model is leveraging a pre-trained VGG19 network as a feature extractor and Support Vector Machines as a classifier.

The model has been trained on a dataset of 9928 images, validated on 2487 images and tested on 5321 images. It has achieved a sample-wise accuracy of 85.6% and a class-wise accuracy of 83.8% on the test set.

Artifacts

Team and role

Team size: 1 person

• I was responsible for the whole project, from model exploration to final implementation.
  

Tech stack