def imgs_to_file(adv_imgs='', pics_root_path=''):
from publicFunctions import load_net_inputs, load_net_params, load_dset_params
import matplotlib.pyplot as plt
npy_name = '/%s_imgs_step%1.1f_linf%d_%dx%d_%s.npy' % (adv_method, step, linf, i_max, j_max, dis_method)
npy_path = 'save_for_load/' + net1 + npy_name
path_white_test_dis_npy = 'save_for_load/distanceADVRetrieval/test_dis_%s.npy' % (net1)
path_black_test_dis_npy = 'save_for_load/distanceADVRetrieval/test_dis_%s.npy' % (net2)
dset_test, dset_database = load_dset_params(job_dataset)
model1, snapshot_path, query_path, database_path = load_net_params(net1)
tmp = np.load(database_path)
_, code, multi_label = tmp['arr_0'], tmp['arr_1'], tmp['arr_2']
test_dis_white, test_dis_black = get_test_dis(path_white_test_dis_npy, path_black_test_dis_npy)
test_true_id_x, test_true_label_y = choose_index_by_dis_method(dis_method, test_dis_white, test_dis_black,
max_dis=18, min_dis=12)
id_size = test_true_id_x.shape[0]
print('id size:', id_size)
i_index_set = np.arange(0, id_size, id_size / (i_max))[:i_max]
j_index_matrix = get_unique_index(code, multi_label, j_max)
adv_imgs = np.load(npy_path)
pics_root_path = './save_for_load/pics/%s_imgs_step%1.1f_linf%d_%dx%d_%s/' %(adv_method, step, linf, i_max, j_max, dis_method)
if not os.path.exists(pics_root_path):
os.makedirs(pics_root_path)
for i in range(i_max):
for j in range(j_max):
print('i,j:',i,j)
file_name_full = pics_root_path + '/' + 'i%s_j%s.jpg'%(str(i), str(j))
img_array = np.moveaxis(adv_imgs[i, j], 0, -1)
plt.imsave(file_name_full, img_array)
return
def main_func():
from publicFunctions import load_net_inputs, load_net_params, load_dset_params
npy_name = '/%s_imgs_step%1.1f_linf%d_%dx%d_%s.npy' % (adv_method, step, linf, i_max, j_max, dis_method)
npy_path = 'save_for_load/' + net1 + npy_name
path_white_test_dis_npy = 'save_for_load/distanceADVRetrieval/test_dis_%s.npy'%(net1)
path_black_test_dis_npy = 'save_for_load/distanceADVRetrieval/test_dis_%s.npy'%(net2)
dset_test, dset_database = load_dset_params(job_dataset)
model1, snapshot_path, query_path, database_path = load_net_params(net1)
tmp = np.load(database_path)
_, code, multi_label = tmp['arr_0'], tmp['arr_1'], tmp['arr_2']
test_dis_white, test_dis_black = get_test_dis(path_white_test_dis_npy, path_black_test_dis_npy)
test_true_id_x, test_true_label_y = choose_index_by_dis_method(dis_method, test_dis_white, test_dis_black,
max_dis=18, min_dis=12)
id_size = test_true_id_x.shape[0]
print('id size:',id_size)
i_index_set = np.arange(0, id_size, id_size / (i_max))[:i_max]
inputs_ori_tensor = torch.stack([dset_test[test_true_id_x[i_index_set[i]]][0] for i in range(i_max)])
j_index_matrix = get_unique_index(code, multi_label, j_max)
adv_imgs = np.load(npy_path)
ori_imgs = inputs_ori_tensor.cpu().numpy()
# imgs_test = np.load('save_for_load/imgs_test.npy')
# target_img_mat = np.load('save_for_load/target_imgs.npy')
target_img_mat = get_target_imgs(j_index_matrix, test_true_label_y, i_index_set, dset_database)
from myRetrieval import get_img_num_by_class_from_img_batch, get_targeted_from_all_class
inputs_targets = Variable(torch.Tensor(target_img_mat).cuda(), requires_grad=True)
img_num_by_class_target = get_img_num_by_class_from_img_batch(inputs_targets, model1, code, multi_label,
threshold=5, batch_size=16)
target_targetedNum_mat = np.zeros([i_max, j_max])
for i in range(i_max):
j_index_set = j_index_matrix[int(test_true_label_y[i_index_set[i]])].astype(int)
label_targeted = np.array([multi_label[j_index_set[j]] for j in range(j_max)])
img_num_target_targeted = get_targeted_from_all_class(img_num_by_class_target[i], label_targeted)
target_targetedNum_mat[i] = img_num_target_targeted
target_targeted_retrieval_num_path = './save_for_load/%s/target_targetedRetrievalNum_%s_%s.npy' % (net1, adv_method, dis_method)
np.save(target_targeted_retrieval_num_path, target_targetedNum_mat)
'''
for i in range(i_max):
for j in range(j_max):
j_index_set = j_index_matrix[int(test_true_label_y[i_index_set[i]])].astype(int)
target_img = target_img_mat[i, j]
#target_result = get_target_retrival_result(model1, )
i, j = 3, 9
ori_img = ori_imgs[i]
adv_img = adv_imgs[i,j]
j_index_set = j_index_matrix[int(test_true_label_y[i_index_set[i]])].astype(int)
label_targeted = np.array([database_label[j_index_set[j]] for j in range(j_max)])
target_label = label_targeted[j]
perturbation_ratio_bound = estimate_subspace_size(adv_img, ori_img, model1, target_label, code, database_label)
'''
return
def get_feature_np(sub_model, inputs):
# convIndex is the index for submodel conv, starting from 1
layer_index_value = layer_index_value_list[net]
model, snapshot_path, query_path, database_path = load_net_params(net)
dset_test, dset_database = load_dset_params(job_dataset)
layer_index = layer_index_value[-2]
sub_model = getConvLayerByIndex(model, layer_index, net)
feature_out = sub_model(inputs)
feature_np = feature_out.cpu().data.numpy()
return 0
def main_backup():
net1 = 'ResNet152'
# convIndex is the index for submodel conv, starting from 1
from .publicVariables import layer_index_value_list
layer_index_value1 = layer_index_value_list[net1]
model1, snapshot_path, query_path, database_path = load_net_params(net1)
dset_test, dset_database = load_dset_params(job_dataset)
layer_index = layer_index_value1[-2]
sub_model1 = getConvLayerByIndex(model1, layer_index, net1)
feature_out1 = sub_model1(inputs)
feature_np1 = feature_out1.cpu().data.numpy()
net2 = 'ResNext101_32x4d'
# convIndex is the index for submodel conv, starting from 1
layer_index_value2 = layer_index_value_list[net2]
model2, snapshot_path, query_path, database_path = load_net_params(net2)
dset_test, dset_database = load_dset_params(job_dataset)
layer_index2 = layer_index_value2[-1]
sub_model2 = getConvLayerByIndex(model2, layer_index2, net2)
feature_out2 = sub_model2(inputs)
feature_np2 = feature_out2.cpu().data.numpy()
his_feature1 = np.histogram(feature_np1[feature_np1 <= 1])
his_feature2 = np.histogram(feature_np2[feature_np2 <= 1])
his_normalized_1 = his_feature1[0].astype(float) / his_feature1[0].sum()
his_normalized_2 = his_feature2[0].astype(float) / his_feature2[0].sum()
from scipy.stats import wasserstein_distance as emd
emd1_2 = emd(his_normalized_1, his_normalized_2)
emd2_1 = emd(his_normalized_2, his_normalized_1)
print(emd1_2, emd2_1)
np.sum(np.abs(tCodeValue - oCodeValue)) / 2)
return adv
if __name__ == "__main__":
job_dataset = 'imagenet'
job_values = ['mnist', 'cifar10', 'fashion_mnist']
net_values = ['ResNet18', 'ResNet34', 'AlexNet']
net = 'ResNet152'
step = 1.0
linf = 32
adv_method = 'miFGSM'
adv_method_list = ['iFGSM', 'iFGSMDI', 'iFGSMMT', 'iFGSMMTDI']
dset_test, dset_database = load_dset_params(job_dataset)
model, snapshot_path, query_path, database_path = load_net_params(net)
tmp = np.load(database_path)
output, code, multi_label = tmp['arr_0'], tmp['arr_1'], tmp['arr_2']
tmp = np.load(query_path)
output_test, code_test, multi_label_test = tmp['arr_0'], tmp['arr_1'], tmp[
'arr_2']
# set index for the targeted image
index = 4
# Load the advertised img
ad_datapath = '../data/ad_dataset/ads/0/'
datapath_dir = os.listdir(ad_datapath)
