This is a short report for the Stepik Computer Vision course. Two general model architectures - typical ConvNet with max-pooling and ResNet - were applied to the classification task on the CIFAR10 dataset. Deep but narrow ResNet20 was compared with the much larger ResNet18 designed for the ImageNet task. For ResNet20 model hyperparameter search was conducted 1) on the dropout rate of two convolutional layers of each ResNet block (0<p<0.2) and 2) weight decay of the Adam optimizer (5e-3<wd<1e-5).