WangFP-516 / One-Pixel-Shotcut

One-Pixel Shortcut: on the Learning Preference of Deep Neural Networks (ICLR 2023 Spotlight)

One-Pixel Shortcut

Official Pytorch implementation of paper One-Pixel Shortcut: on the Learning Preference of Deep Neural Networks

Accepted to ICLR 2023 as a spotlight paper.

By Shutong Wu, Sizhe Chen, Cihang Xie and Xiaolin Huang

Paper Link: https://arxiv.org/abs/2205.12141

Requirements

Here are the versions of packages we use for the implementation of experiments.

Library	Version
Python	3.7.7
pytorch	1.7.1
torchvision	0.8.2
numpy	1.20.3
tqdm	4.62.2

Run the main One-Pixel Shortcut training and evaluation script

For example, here is the command to train a ResNet-18 on OPS data:

python main.py \
--data_path=the location of your dataset \
--save_path=the saving location and name of this experiment \
--pert=OPS \
--model=RN18 \
--data_aug=Standard \
--sparsity=1 \
--at_pgd_step=0 \

if you want to train a ResNet-18 on CIFAR-10-S data, run:

python main.py \
--data_path=the location of your dataset \
--save_path=the saving location and name of this experiment \
--pert=CIFAR10-S \
--em_path=the location of EM noise file \
--model=RN18 \
--data_aug=Standard \
--sparsity=1 \
--at_pgd_step=0 \

Due to the limitation of file size, we do not include our pre-generated EM noise here. Please see https://github.com/HanxunH/Unlearnable-Examples for details of EM noise generation. Then set em_path to the location where you save the EM noise file.