Skip to content

Latest commit

 

History

History
34 lines (24 loc) · 938 Bytes

README.md

File metadata and controls

34 lines (24 loc) · 938 Bytes
Raw

Prerequisites

  • Python 3.7
  • PyTorch 1.3.1
  • PIL
  • OpenCV

Training & Evaluation

Usage (example):

mkdir logs
mkdir outputs_
bash train_test.sh $GPU_ID $tmp $DATASET_NAME $mode $w1 $w2 $num_works $bsz

$GPU_ID is the lanched GPU ID and w1, w2, temp denote as mu, sigma and temperature respectively. $DATASET_NAME is the dataset name (cifar10, cifar100, stl10), mode represents the model (easy or adnce). $num_works and $bsz are hyperparamters for training:

Commands for reproducing the reported results:

### CIFAR10
bash train_test.sh 0 0.4 cifar10 adnce 0.7 0.5 8 256

### CIFAR100
bash train_test.sh 0 0.3 cifar100 adnce 0.5 1.0 8 256

### STL10
bash train_test.sh 0 0.2 stl10 adnce 0.8 1.0 8 256

Acknowledgements

Part of this code is inspired by leftthomas/SimCLR and chingyaoc/DCL.