Skip to content

A model to classify images from the CIFAR-10 dataset using PyTorch.

Notifications You must be signed in to change notification settings

martinoywa/cifar10-cnn-exercise

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 

Repository files navigation

cifar10-cnn-exercise

A model to classify images from the CIFAR-10 dataset using PyTorch. The model uses 3 Convolutional Layers, Maxpooling Layers, 3 Fully Connected(Linear Layers) and Dropout Layers with 50% probability.

test loss and accuracy for all image classes

Test Loss: 0.787522

Test Accuracy of airplane: 76% (760/1000)
Test Accuracy of automobile: 87% (871/1000)
Test Accuracy of bird: 60% (600/1000)
Test Accuracy of cat: 56% (564/1000)
Test Accuracy of deer: 69% (692/1000)
Test Accuracy of dog: 61% (612/1000)
Test Accuracy of frog: 81% (815/1000)
Test Accuracy of horse: 78% (785/1000)
Test Accuracy of ship: 87% (871/1000)
Test Accuracy of truck: 83% (831/1000)

overall accuracy

Test Accuracy (Overall): 74% (7401/10000)

model's weaknesses and how might they be improved

model is doing really poorly in classifying animals especially those with varying colors like cats and birds. Create another Convolutional layer that classifies finer details such as colors. A lower learning rate also, so that there's is faster convergence. Augmentation might also improve the model.

sample predictions

sample predictions

About

A model to classify images from the CIFAR-10 dataset using PyTorch.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published