Using Autoencoders for feature extraction on CIFAR-10 dataset and passing the encoded features to image classifier network. Two types of architecture models have been trained.
-
End to end model
This model combines encoder module of autoencoder with image classifer by avoiding the decoder module of autoencoder. Thus, the generated code of encoder is passed to image classifier.
-
Seperate-training model
In this model, the autoencoders are trained first and then the encoded data is fed to image classifier network.
Further, for the purpose of comparison, we have used simple convolutional neural network model where the CNN architecture is same as architecture used for image classifer network by the other models.
The report of this assignment can be downloaded at Report.pdf
To run end-to-end model , type
python3 endtoend.py
a. Confusion Matrix Heat Map
b. Random Image and its predicted Class
\
c. Model accuracy and Loss
To run seperate-training model , type
python3 seperate-classifier.py
a. Confusion Matrix Heat Map
b. Random Image and its predicted Class
c. Model accuracy and Loss
To run simplre-CNN model , type
python3 classic-conv.py
a. Confusion Matrix Heat Map
b. Random Image and its predicted Class
c. Model accuracy and Loss
\
Comparision of all three models
a. Scatter Plot
