Skip to content

Latest commit

 

History

History
49 lines (35 loc) · 4.56 KB

fig5_targeted.md

File metadata and controls

49 lines (35 loc) · 4.56 KB
Raw

Figure 5 - CDFS for targeted attacks

The following calls can be used to reproduce the results from Fig. 5 of the paper, representing the cumulative distribution functions of several targeted black-box attacks, with four surrogate networks ({VGG-19, ResNet-34, DenseNet-121, MobileNetV2}). The results for GFCS and SimBA-ODS are obtained with the script GFCS_main.py, which belongs to this repository.

If also reproducing the Square Attack results, make sure that that repo has been configured as specified in fig2_untargeted.md.

Perform attacks using four surrogates

GFCS

python GFCS_main.py --model_name vgg16_bn --smodel_name vgg19_bn resnet34 densenet121 mobilenet_v2 --data_index_set vgg16_bn_batch0 --GFCS --num_sample 1000 --num_step 30000 --norm_bound 12.269 --step_size 2.0 --output experimental_results/2022/targeted/4surr/targeted_GFCS_vgg16target_r34v19d121mobv2surr_vgg16_bn_batch0_30000_12.269_2.0.pt --targeted
python GFCS_main.py --model_name resnet50 --smodel_name vgg19_bn resnet34 densenet121 mobilenet_v2 --data_index_set resnet50_batch0 --GFCS --num_sample 1000 --num_step 30000 --norm_bound 12.269 --step_size 2.0 --output experimental_results/2022/targeted/4surr/targeted_GFCS_res50target_r34v19d121mobv2surr_resnet50_batch0_30000_12.269_2.0.pt --targeted
python GFCS_main.py --model_name inception_v3 --smodel_name vgg19_bn resnet34 densenet121 mobilenet_v2 --data_index_set inceptionv3_batch0 --ODS  --special_margin_gradient_option --revert_to_ODS_when_stuck --net_specific_resampling --num_sample 1000 --num_step 30000 --norm_bound 16.377 --step_size 2.0 --output experimental_results/2022/targeted/4surr/targeted_GFCS_incep3target_r34v19d121mobv2surr_interp_inceptionv3_batch0_30000_16.377_2.0.pt --targeted

SimBA-ODS

python GFCS_main.py --model_name vgg16_bn --smodel_name vgg19_bn resnet34 densenet121 mobilenet_v2 --data_index_set vgg16_bn_batch0 --ODS --num_sample 1000 --num_step 30000 --norm_bound 12.269 --step_size 2.0 --output experimental_results/2022/targeted/targeted_4surr/SimbaODS_vgg16target_r34v19d121mobv2surr_vgg16_bn_batch0_30000_12.269_2.0.pt --targeted
python GFCS_main.py --model_name resnet50 --smodel_name vgg19_bn resnet34 densenet121 mobilenet_v2 --data_index_set resnet50_batch0 --ODS --num_sample 1000 --num_step 30000 --norm_bound 12.269 --step_size 2.0 --output experimental_results/2022/targeted/4surr/targeted_SimbaODS_res50target_r34v19d121mobv2surr_resnet50_batch0_30000_12.269_2.0.pt --targeted 
python GFCS_main.py --model_name inception_v3 --smodel_name vgg19_bn resnet34 densenet121 mobilenet_v2 --data_index_set inceptionv3_batch0 --ODS --net_specific_resampling --num_sample 1000 --num_step 30000 --norm_bound 16.377 --step_size 2.0 --output experimental_results/2022/targeted/4surr/targeted_SimbaODS_incep3target_r34v19d121mobv2surr_interp_inceptionv3_batch0_30000_16.377_2.0.pt --targeted

SimBA-ODS with --step-size 0.2

python GFCS_main.py --model_name vgg16_bn --smodel_name vgg19_bn resnet34 densenet121 mobilenet_v2 --data_index_set vgg16_bn_batch0 --ODS --num_sample 1000 --num_step 30000 --norm_bound 12.269 --step_size 0.2 --output experimental_results/2022/targeted/4surr/targeted_SimbaODS_vgg16target_r34v19d121mobv2surr_vgg16_bn_batch0_30000_12.269_0.2.pt --targeted
python GFCS_main.py --model_name resnet50 --smodel_name vgg19_bn resnet34 densenet121 mobilenet_v2 --data_index_set resnet50_batch0 --ODS --num_sample 1000 --num_step 30000 --norm_bound 12.269 --step_size 0.2 --output experimental_results/2022/targeted/4surr/targeted_SimbaODS_res50target_r34v19d121mobv2surr_resnet50_batch0_30000_12.269_0.2.pt --targeted 
python GFCS_main.py --model_name inception_v3 --smodel_name vgg19_bn resnet34 densenet121 mobilenet_v2 --data_index_set inceptionv3_batch0 --ODS --net_specific_resampling --num_sample 1000 --num_step 30000 --norm_bound 16.377 --step_size 0.2 --output experimental_results/2022/targeted/4surr/targeted_SimbaODS_incep3target_r34v19d121mobv2surr_interp_inceptionv3_batch0_30000_16.377_0.2.pt --targeted

Generate plots data and figures

Prepare results

python plot_results.py --input experimental_results/2022/targeted/4surr

Plot results

python generate_results_summary.py --expm_json plots_setup/2022/targeted/4surr/vgg16.json --save_to plots_paper/2022/targeted/4surr/vgg16.pdf
python generate_results_summary.py --expm_json plots_setup/2022/targeted/4surr/resnet50.json --save_to plots_paper/2022/targeted/4surr/resnet50.pdf
python generate_results_summary.py --expm_json plots_setup/2022/targeted/4surr/incep3.json --save_to plots_paper/2022/targeted/4surr/incep3.pdf