This study implemented the classic image classification with various CNN (convolutional neural network) models on CIFAR10 dataset.
This study applied classic AlexNet, VGG11 and ResNet18 models and compared the difference between models.
For further experiment, learning rate decay, different optimizer methods, fine-tuning are applied. After all experiments, without fine-tuning, the best final model is ResNet18, with test accuracy 86% and with fine-tuning, ResNet18 has test accuracy as 94%.
Detailed discussion can be found in CNN_review.pdf