Classifying images using Convolutional Neural Networks

Nov 6, 2021
Code
The best model for the test data

Decription:

In this project, We investigate several configurations of CNNs to classify input images.

Dataset

For this task, we’ve used Fashion MNSIT which contains 10 classes of data for a classification task.

Training

We’ve trained Network with several Configurations to find the best setting for our task. We have checked for these settings:

  • MLP with a different number of hidden layers
  • MLP with different Activation functions
  • Convolutional Neural Networks
  • Convolutional Neural Networks with DropOut
  • Convolutional Neural Networks with DropOut and Batch normalization

Results

Accuracy for different settings mentioned above are in the below table:

Model Accuracy%
CNN + fully connected 90,63
CNN + fully connected + Pooling + Batch normalization 90,17
CNN + fully connected + Pooling + CNN_Dropout(0.3,0.3,0.5) 90,56
CNN + fully connected + Pooling + CNN_Dropout(0.2,0.2,0.2) 90,89
CNN + fully connected + Pooling + Batchnormalization + CNN_Dropout(0.2,0.2,0.2) 91,04
Past Deep Learning
Zahra Habibzadeh
Zahra Habibzadeh
Research Assistant in Artificial Intelligence and Robotics

My research interests include Computational Social Science, Natural Language Processing, and Data Mining.