-
Notifications
You must be signed in to change notification settings - Fork 0
/
eval_patches.sh
61 lines (46 loc) · 2.15 KB
/
eval_patches.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
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
#!/bin/bash
#SBATCH --job-name="eval_patches"
#SBATCH --ntasks=1
#SBATCH --gres=gpu:1 # use 1 GPU
#SBATCH --cpus-per-gpu=1
#SBATCH -o logs/eval_patches.o%j
#SBATCH --mem-per-gpu=30G ## GMA: 2.7jGB, FlowFormer 6,3GB
# export CUDA_VISIBLE_DEVICES=1
source $PATCHATTACK_HOME/venv/patch-attack/bin/activate
cd $PATCHATTACK_HOME
# model to evaluate. net is first argument, weights is second
net=FlowNetC
weights=models/_pretrained_weights/FlowNet2-C_checkpoint.pth.tar #raft-kitti.pth # models/_pretrained_weights/pwc_net_chairs.pth.tar #FlowNet2-C_checkpoint.pth.tar #raft-kitti.pth #
# eval best n patches (with the same parameter combinations)
# n=1 # currently head -n ${n} is commented out
# defense on which patches were trained
for defense in none lgs ilp; do
# folder where patches are stored
folder=results/${net}_${defense}_evals/best-patches/
# folder=results/Manual_${defense}_evals/best-patches/
# number of images to evaluate on
N=200
# fixed parameters
exp_name=eval_${net}_${defense}
# exp_name=eval_Manual_${net}_${defense}
s=15
t=0.15
o=8
k=16
r=5
# first n patches from folder
for patch in $(ls ${folder}) #| head -n ${n}) # | grep png)
do
echo "patch: $patch"
## no defense
python evaluate_patch_withDefense.py --dataset Kitti15 --dataset_stage training --net $net -w $weights --patch_name ${folder}/${patch} --defense none --custom_experiment_name ${exp_name}_none --n $N &
sleep 5 # wait for experiment to be created
# LGS
python evaluate_patch_withDefense.py --dataset Kitti15 --dataset_stage training --net $net -w $weights --patch_name ${folder}/${patch} --defense lgs --s $s --t $t --custom_experiment_name ${exp_name}_lgs --n $N &
sleep 5 # wait for experiment to be created
# ILP
python evaluate_patch_withDefense.py --dataset Kitti15 --dataset_stage training --net $net -w $weights --patch_name ${folder}/${patch} --defense ilp --r $r --s $s --t $t --k $k --o $o --custom_experiment_name ${exp_name}_ilp --n $N
sleep 5 # wait for experiment to be created
done
done
exit