04 / Computer vision · Deep learning
ResNet18 & VGG13
ResNet18 and VGG13 take on the same handwritten digits. A comparison built in PyTorch.
The question
How do two different network architectures approach the same image classification task?
The approach
I trained ResNet18 and VGG13 on MNIST using PyTorch. Keeping the dataset the same focuses the comparison on the architectures, including ResNet's residual connections and VGG's sequential layers.
Project scope
- MNIST image classification
- Two PyTorch model implementations
- Architecture comparison & evaluation
How it works
A shared dataset
Both networks use MNIST, a dataset of handwritten digits. The task is to assign an input image to a digit class.
Different paths through a network
ResNet18 uses residual connections. VGG13 uses a sequential stack of layers. The project puts both architectures to work on the same task.
Training and evaluation
I implemented and evaluated both networks in PyTorch. The scope spans the model implementation, training and prediction stages, rather than using a hosted image-classification service.
The build
What I built
An image-classification comparison with two independently implemented networks. My work covered training and evaluating both models on handwritten digits, connecting the architecture of each network to the same classification task.
Next project / 05