Implementation of the ResNet model in PyTorch. Based on the architecture from the paper: https://arxiv.org/abs/1512.03385.
Inspired by the torchvision implementation: https://github.com/pytorch/vision.
Trained on CIFAR-10 dataset: https://www.cs.toronto.edu/~kriz/cifar.html.
epochs: 90
batch size: 128
learning rate: 0.1 (divided by 10 when error plateaus)
optimizer: SGD (weight decay 1e-4, momentum 0.9)
loss function: Cross entropy
| ResNet-34 | ResNet-50 | ||
|---|---|---|---|
| # of Trainable Params | 21.3M | 23.5M | |
| Test Accuracy | 87.8% | 86.2% |