This residual network is trained on CIFAR-10 datasets, and eventually reached 94.44% accuracy.

Run the notebook "TrainingScript" to retrain the network on CIFAR-10 and weight file will be saved as "project1_model.pt". Training 200 epoches can be time consuming, but accuracy stabalizes after 120+ epoches in our run (reference graph below), so feel free to pick your epoches.

Run the notebook "TestingScript", you can choose which trained weight by changing the path in

resnet.load_state_dict(torch.load('./project1_model.pt'))

• TrainingScript.ipynb

  jupyter notebook to reproduce training process and graphs

• TestingScript.ipynb

  jupyter notebook to validate trained weight

• project1_model.py

  definition of the ResNet-18 network

• project1_model.pt

  trained weight for the ResNet-18 network