The goal of this project is to create 2 types of Neural Network models:
- An image auto-encoder model.
- An image classification model.
The auto-encoder model consists of two different types of layers: the encoding and the decoding layers. The program created is a handy interface in order for the user to insert different values of several hyper-parameters and see the behavior of each model, with ultimate goal to chose one model that best handles the dataset given.
The classification model aims to classifying images in a category. To do so, it uses a pre-trained auto-encoder model, by taking advantage of its encoding layers, which are then connected to a fully connected layer, and then to an output one, aiming to the best possible classification of the images. Once again, multiple models can be trained, in order for the user to decide the best for the training set needs, and then, the best one will be used in order to predict the test dataset.
In order to run the Autoencoder model, you should navigate to the directory Autoencoder, and run the file autoencoder.py, as following:
python autoencoder.py −d <trainingset>
In order to run the classifier model, you should navigate to the directory classifier, and run the file classification.py, as following:
python classification.py −d <trainingset> −dl <traininglabels> −t <testset> − tl <testlabels> −model <autoencoderh5>
Specifics of our code implementation can be found under the README.pdf file.
Some sample results from our experiments in the models are the following:
Error Plot:
Accuracy Plot:
We also created a high dimensional plot in order to visualze the accuracy and the loss for our best models, using HiPlot.
Accuracy Plot:
Error Plot:
The hiplot created for our best classification models is the following:
The result of the predictions from our best model(accuracy 99%), are the following
This project is licensed under the MIT License - see the LICENSE file for details
