This repo is for the DL mini-project1. In this project, a ResNet-18 architecture is designed to maximize the accuracy on the CIFAR-10 dataset under a constraint of less than 5M trainable parameters. ResNet hyper-parameters are tuned. In model training, some strategies are used, such as data augmentation, optimizer, regularization scheme. Model 1 to model 7 are the models we used for experiment.\ We saved our final model in DLProject1/model7. To train the model, please run "python load_test.py" under the path of DLProject1/model7.