This small project was performed as an assignment for the course of Artificial Neural Networks & Deep Learning and it's part of the exam. The task was to develop a model for Image Classification and to train it to distinguis between 14 classes of leaves, as in the example below.
The dataset we used was composed of 17728 images, but training with only those resulted in a poor accuracy on the private test set. We used the ImageDataGenerator class to perform data augmentation.
The ultimate model we submitted reached an accuracy of 92.08% on the private test set and was obtained using the technique of Transfer Learning; the model was then fine-tuned for two times, with a progressively smaller learning rate. Find more accurate description on the development and the specific details of the models on the attached document.
Group Name: Gamma