0.0.1

fawkes_dev/README.md

@@ -46,6 +46,13 @@ To evaluate the cloak, run `python3 fawkes/eval_cloak.py --gpu 0 --cloak_data PA
 The code will print out the tracker model accuracy on uncloaked/original test images of the protected user, which should be close to 0. 
 
 
+#### Exisiting Feature extractors
+
+We shared three different feature extractors under feature_extractors/
+1. low_extract.h5: trained on WebFace dataset with DenseNet architecture. 
+2. mid_extract.h5: VGGFace2 dataset with DenseNet architecture. Trained with PGD adversarial training for 5 epochs. 
+3. high_extract.h5: WebFace dataset with DenseNet architecture. Trained with PGD adversarial training for 20 epochs. 
+
 ### Citation
 ```
 @inproceedings{shan2020fawkes,

fawkes_dev/config.py

@@ -3,8 +3,8 @@ import json
 import os
 
 DATASETS = {
+    "pubfig": "../data/pubfig",
     "scrub": "/home/shansixioing/cloak/fawkes/data/scrub/",
-    # "vggface1": "/mnt/data/sixiongshan/data/vggface/",
     "vggface2": "/mnt/data/sixiongshan/data/vggface2/",
     "webface": "/mnt/data/sixiongshan/data/webface/",
     "youtubeface": "/mnt/data/sixiongshan/data/youtubeface/keras_flow_data/",

fawkes_dev/differentiator.py

@@ -207,8 +207,7 @@ class FawkesMaskGeneration:
         # we're creating
         start_vars = set(x.name for x in tf.global_variables())
         self.learning_rate_holder = tf.placeholder(tf.float32, shape=[])
-        # optimizer = tf.train.AdadeltaOptimizer(self.learning_rate_holder)
-        optimizer = tf.train.AdamOptimizer(self.learning_rate_holder)
+        optimizer = tf.train.AdadeltaOptimizer(self.learning_rate_holder)
 
         self.train = optimizer.minimize(self.loss_sum,
                                         var_list=[self.modifier])

fawkes_dev/prepare_feature_extractor.py

@@ -44,7 +44,7 @@ def main():
     for k, v in path2emb.items():
         path2emb[k] = normalize(v)
 
-    pickle.dump(path2emb, open("../feature_extractors/embeddings/{}_emb_norm.p".format(args.feature_extractor), "wb"))
+    pickle.dump(path2emb, open("../feature_extractors/embeddings/{}_emb.p".format(args.feature_extractor), "wb"))
 
 
 def parse_arguments(argv):

fawkes_dev/protection.py

@@ -83,9 +83,9 @@ def parse_arguments(argv):
                         help="name of the feature extractor used for optimization",
                         default="webface_dense_robust_extract")
     parser.add_argument('--th', type=float, default=0.007)
-    parser.add_argument('--sd', type=int, default=1e5)
+    parser.add_argument('--sd', type=int, default=1e9)
     parser.add_argument('--protect_class', type=str, default=None)
-    parser.add_argument('--lr', type=float, default=0.1)
+    parser.add_argument('--lr', type=float, default=1)
 
     return parser.parse_args(argv)
 

fawkes_dev/utils.py

@@ -13,7 +13,7 @@ from keras.models import Model
 from keras.preprocessing import image
 from keras.utils import to_categorical
 from sklearn.metrics import pairwise_distances
-
+# from keras.utils import get_file
 
 def clip_img(X, preprocessing='raw'):
     X = reverse_preprocess(X, preprocessing)
@@ -172,6 +172,7 @@ def imagenet_reverse_preprocessing(x, data_format=None):
     return x
 
 
+
 def build_bottleneck_model(model, cut_off):
     bottleneck_model = Model(model.input, model.get_layer(cut_off).output)
     bottleneck_model.compile(loss='categorical_crossentropy',