/
run.sh
30 lines (27 loc) · 793 Bytes
/
run.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
#!/bin/bash
# Fixed Params
PRETRAINED_DATASET="imagenet"
DATASET="places365"
EPSILON=0.03922
LOSS_FN="bounded_logit_fixed_ref"
CONFIDENCE=10
BATCH_SIZE=32
TARGET_CLASS=150
LEARNING_RATE=0.005
NUM_ITERATIONS=2000
WORKERS=4
NGPU=1
SUBF="imagenet_targeted"
TARGET_NETS="alexnet googlenet vgg16 vgg19 resnet152"
for target_net in $TARGET_NETS; do
python3 train_uap.py \
--dataset $DATASET \
--pretrained_dataset $PRETRAINED_DATASET --pretrained_arch $target_net \
--target_class $TARGET_CLASS --targeted \
--epsilon $EPSILON \
--loss_function $LOSS_FN --confidence $CONFIDENCE \
--num_iterations $NUM_ITERATIONS \
--batch_size $BATCH_SIZE --learning_rate $LEARNING_RATE \
--workers $WORKERS --ngpu $NGPU \
--result_subfolder $SUBF
done