Requirements and Installation

  • A computer running macOS or Linux
  • For training new models, you’ll also need a NVIDIA GPU and NCCL
  • Python version 3.6
  • A PyTorch installation

FILE description

  • ./checkpoint : Stored with trained models
    ResNet18 : Stored baseline best model of ResNet18
    ResNet18_v2 : Stored the best model of ResNet18 using the modified mixup, including the results obtained by training 200 epochs by default and training 1000 epochs respectively
    DenseNet190 : Stored baseline best model of DenseNet190
    DenseNet190_v2 : Stored the best model of DenseNet190 using the modified mixup

  • ./results : Stored training log
    ResNet18_baseline
    ResNet18_v2_epoch200
    ResNet18_v2_epoch1000
    DenseNet190
    DenseNet190_v2_epoch200

  • ./mixup.py : original mixup function

  • ./mixup_v2.py : modified mixup function

  • ./train.py
    (Modify line 25 during training. If you use the original mixup, change it to import mixup as mp
    Use the modified mixup to import mixup_v2 as mp)

Training

Use python train.py to train a new model.
Here is an example setting:

$ CUDA_VISIBLE_DEVICES=0 python train.py --lr=0.1 --seed=20170922 --decay=1e-4

GitHub

View Github