Please first make sure you have properly installed Cuda on your device. The requirements.txt lists all the requirements for this implementation. You can install all the requirements by running
pip3 install -r requirements.txt
in the prompt under your code folder.
To train a ResNet18 on CIFAR-10, use the following command (in a Windows environment):
python cifar_trainer.py --arch resnet18 --coeff -1.0 --dataset cifar10 --save-dir folder_name --gpu 0
To check the test error and disagreement rate, please run the following command:
.\eval.py --path1 folder_path_of_model_1 --path2 folder_path_of_model_2 --gpu 0 --output_name test_name
Please note that the first value displayed after running the command is the test accuracy in %.
The author would like to thank Taejong Joo, a fellow first-year IEMS PhD for valuable discussions in the course of implementing ResNet18 on CIFAR-10. The code implementation is based on the code from his GitHub site: https://tjoo512.github.io/.