advPatch-pytorch

code for generating adversarial patches

Required packages

pip install pytorch-msssim

Installation

Clone the project with the submodules.

git clone --rescursive URL

This repo requires Python >= 3.6. To get dependent packages, you can install the required packages in requirement.txt via

pip install -r requirement.txt

In order to use the object detectors from SSD or Detectrons, you will need to install them from the submodule

To install SSD, more details can be found here.

cd REPO/detector/SSD
pip install -e . 

To install Detectron2, more details can be found here

cd REPO/detector/detectron2
pip install -e .

#Usage

1 Learning STN with Generator

1. Training

python train_patchTransformer.py --config configs/config_patchTransformer.yaml --logdir STN-results --dataset neu_color \
   --datadir ../../adv_data/neu_data --epochs 600 --STN tps --learnableSTN --use_LCT --LightingCT gen --batch_size 72`

2. Evaluation

python train_patchTransformer.py --config configs/config_patchTransformer.yaml --logdir thinklab-STN-results --dataset neu_color \
  --datadir ../../adv_data/neu_data --epochs 600 --STN tps --learnableSTN --use_LCT --LightingCT gen --batch_size 60 \
  --patch_transformer_path thinklab-STN-results/PT_neu_color_STN_resnet18_ds128_fc256_tps_bounded20x10_gen_p256_L1Mask_bs60_e600/model_best.pth.tar \
  --visualize --evaluate

--visualize: save intermediate results into a folder 'vis_output' under 'patch_transformer_path' --val_list_file: specify which subset to be evaluated: train or validation

2 Learning Printer Color (PCT) and Lighting Transformation (LCT)

1. Training

python train_patchTransformer.py --config configs/config_patchTransformer.yaml --logdir PatNet-results --dataset neu_color \
--datadir ../../adv_data/neu_data --epochs 600 --STN tps --use_PCT --PrinterCT PCTLinear --use_LCT --LightingCT cc --batch_size 72 \
--pretrained STN-results/PT_neu_color_STN_resnet18_ds128_fc256_tps_bounded20x10_gen_bs72_e600/model_best.pth.tar 

3 Learning Adversarial Attack Model

1. Training

python train_advPatch.py --config configs/config_advPatch.yaml --logdir AdvNet-results --dataset neu_color --datadir ../../adv_data/neu_data \
--epochs 600 --STN tps --use_PCT --PrinterCT PCTLinear --use_LCT --LightingCT cc --batch_size 72  \
--patch_transformer_path PatNet-results/PatNet/PT_neu_color_fixedSTN_blur6_resnet18_ds128_fc256_tps_bounded20x10_PCTLinear_cc_alexnet_bs72_e600_pretrained_nopctloss_blur/model_best.pth.tar

2. Evaluation

python train_advPatch.py --config configs/config_advPatch.yaml --logdir AdvNet-results --dataset neu_color --datadir ../../adv_data/neu_data \
--epochs 600 --STN tps --use_PCT --PrinterCT PCTLinear --use_LCT --LightingCT cc --batch_size 72  \
--patch_transformer_path PatNet-results/PatNet/PT_neu_color_fixedSTN_blur6_resnet18_ds128_fc256_tps_bounded20x10_PCTLinear_cc_alexnet_bs72_e600_pretrained_nopctloss_blur/model_best.pth.tar\
--evaluate