[Python] [Pytorch]
Initiate training with designated parameters (epoch, batch, optimizer, loss function, etc)
Finish training & testing with plot summary of loss & accuracy rate by epochs
I created a framework that can be used for training & evaluating pytorch models that classifies images into different classes. Then I tried improving the model performance using this framework and also attempted to create my own model.
The datasets I used for this projects are *MNIST, CIFAR10, CIFAR100*
I started from making a simple FCL (Fully-Connected-Layer) model for a MNIST dataset. Then I tried changing the model architechture by changing the number of nodes in the hidden layers and the depth of the model. Plus I tried using different optimizers (learning rate, momentum), loss functions, activation functions to improve the model's accuracy.
Then I tried mimicking models created by others like AlexNet and ResNet by scaling the parameters down to match the datasets I wanted. ( AlexNet, ResNet uses [224x224] images while images in my datasets are [28x28] or [32x32] )
The log for training results can be found underneath
I was able to train and test the nueral networks using Pytorch, which is an machine-learning library for Python. Detailed information about the theories, tools, and code can be found in my Github Repository here → https://github.com/sunghokim128/Computer-Vision-with-Pytorch, check it out!