The Evaluation code of NARes
-
Put dataset (CIFAR-10, CIFAR-10-C) to
datasets/:datasets ├── cifar10 │ └── cifar-10-batches-py └── cifar10c -
Put model weights and config file to
models_home/:models_home └── arch_170 ├── arch_170.yaml └── checkpoints └── arch_170_best.pth -
Install packages from
requirements.txt
model=arch_2333
#Base Linf
python ./eval.py --model-path ./models_home --log-path ./attack_log --load-best-model --model ${model} --seed 0
#Base L2
python ./eval.py --model-path ./models_home --log-path ./attack_log --load-best-model --model ${model} --seed 0 --norm L2
#AA-Linf
python ./eval.py --model-path ./models_home --log-path ./attack_log --load-best-model --model ${model} --seed 0 --attack-choice AA --aa-type Compact
#AA-L2
python ./eval.py --model-path ./models_home --log-path ./attack_log --load-best-model --model ${model} --seed 0 --attack-choice AA --aa-type Compact --norm L2Evalute models by a text file with arch_id list:
# run on GPU0 with Base-Linf and seed 42
./run.sh 0 arch_list.txt base Linf 42
# run on GPU3 with AA-Linf and seed 22
./run.sh 3 arch_list.txt aa Linf 22python ./eval_corruption_level.py --model-path ./models_home --log-path ./attack_log --load-best-model --model arch_2333
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent