Skip to content

athinnyun/UVA_Landmark_Classifier

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 

Repository files navigation

UVA Landmark Classifier

In this project, I trained and evaluated three different CNN-based models via transfer learning to classify different landmarks at the University of Virginia. The first model was based on ResNet50 and it achieved a test accuracy of 85.11%. The second one was based on InceptionV3 and it achieved a test accuracy of 74.61%. The final one was based on Xception and it achieved a test accuracy of 86.94%, the highest of all three. For all three I used the same method of training, using a SGD optimizer with a fairly high learning rate to train just the output layer and then an Adadelta optimizer with default values to train the whole net. I tried a few different optimizers and this configuration worked well for me. Additionally, my initial models suffered from bad overfitting so for each model I applied some Elastic Net regularization to the output layer and included a Dropout layer with a rate of 40% just before the fully connected layer. Overall these are pretty good results but I would like to do further experimentation to achieve a higher accuracy. Credits to Rich Nguyen for providing the dataset and preprocessing code.

UVA Landmarks

Classifier Accuracy

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published