The CNN architecture that we have chosen for this task is ResNet-18. The procedure consisted of first preprocessing the data, and then training the model and tweaking the hyperparameters. For the preprocessing part, the images were resized to be smaller in order to reduce training time. The images were read into numpy arrays and the pixel values were normalized to improve stability. Following that, the labeled data was split into training and testing datasets, which were further split into mini batches for stochastic gradient descent, and the training began.
The CNN architecture that we have chosen for this task is ResNet-18. The procedure consisted of first preprocessing the data, and then training the model and tweaking the hyperparameters. For the preprocessing part, the images were resized to be smaller in order to reduce training time. The images were read into numpy arrays and the pixel values were normalized to improve stability. Following that, the labeled data was split into training and testing datasets, which were further split into mini batches for stochastic gradient descent, and the training began.