An implementation of the CIFAR-10 part of ResNet in PyTorch. For more Information on this network and training parameters, please refer to the original paper here.
Training residual and non-residual networks for all n = 3, 5, 7 and 9:
python train.py
Training a list of specific networks:
python train.py 3,True 8,False will train a residual network for n=3 and a non-residual network for n=8.
The code successfully replicates the main results: Increasing depth beyond n=3 will result in an increase in error rate for the non-residual network and a decrease in error rate for the residual network. Compare with Figure 6 in the original paper.
Pre-trained models are available in the models folder. The naming scheme is analogous to the usage above.