martinoywa / cifar10-cnn-exercise

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

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