def compute_distances(self, features: List[Dict[str, np.array]]):
query_feature = features[self.query_im_path]
pos_feature = features[self.pos_im_path]
neg_features = [features[path] for path in self.neg_im_paths]
self.pos_dist = vector_distance(query_feature,
pos_feature,
method=self.distance_method)
self.neg_dists = np.array(
[vector_distance(query_feature, f) for f in neg_features])
def test_vector_distance():
vec1 = np.array([-1, 0, 1.0])
vec2 = np.array([1, -6.2, 2])
assert vector_distance(vec1, vec2, "l2", l2_normalize=False) == approx(
distance.euclidean(vec1, vec2))
vec1 = vec1 / np.linalg.norm(vec1, 2)
vec2 = vec2 / np.linalg.norm(vec2, 2)
assert vector_distance(vec1, vec2,
"l2") == approx(distance.euclidean(vec1, vec2))
def test_compute_distances():
query_feature = [-1, 0.2, 2]
feature_dict = {"a": [0, 3, 1], "b": [-2, -7.2, -3], "c": [1, 2, 3]}
distances = compute_distances(query_feature, feature_dict)
assert len(distances) == 3
assert distances[1][0] == 'b'
assert vector_distance(query_feature,
feature_dict["b"]) == approx(distances[1][1])
def compare_sig(app_img, device_img):
start_time = time.time()
signature_class_1 = None
signature_class_2 = None
if app_img:
signature_class_1 = classify_learn.predict(app_img)[1]
signature_class_1 = str(signature_class_1.item())
if device_img:
signature_class_2 = classify_learn.predict(device_img)[1]
signature_class_2 = str(signature_class_2.item())
result = ""
similarity_score = -1
if (app_img and device_img):
if (config['sig_config']['similarity_response'][signature_class_1] !=
config['sig_config']['success_case'] and
config['sig_config']['similarity_response'][signature_class_2]
!= config['sig_config']['success_case']):
result = config['sig_config']['similarity_response'][
signature_class_1] if config['sig_config'][
'similarity_response'][signature_class_1] != config[
'sig_config']['success_case'] else config['sig_config'][
'similarity_response'][signature_class_2]
if (result == ""):
featurizer.features = None
similarity_learn.predict(app_img)
app_emb = featurizer.features[0][:]
assert len(app_emb) > 1
featurizer.features = None
similarity_learn.predict(device_img)
device_emb = featurizer.features[0][:]
assert len(device_emb) > 1
featurizer.features = None
similarity_score = vector_distance(app_emb, device_emb)
if similarity_score <= config['sig_config']['accept_threshold']:
result = config['sig_config']['success_case']
elif similarity_score >= config['sig_config']['deny_threshold']:
result = config['sig_config']['fail_case']
else:
result = config['sig_config']['unknown_case']
elif app_img or device_img:
result = app_img and config['sig_config']['similarity_response'][
signature_class_1] or config['sig_config']['similarity_response'][
signature_class_2]
else:
return {"status": "error", 'description': "no files"}
print("compare signature --- %s seconds ---" % (time.time() - start_time))
return {
"status":
result,
"similarity":
str(similarity_score),
"app_class":
signature_class_1
and config['sig_config']['classify_response'][signature_class_1] or "",
"device_class":
signature_class_2
and config['sig_config']['classify_response'][signature_class_2] or ""
}
def compare_sig(app_img, device_img):
start_time = time.time()
signature_class_1 = None
signature_class_2 = None
if not isinstance(app_img, NoneType):
signature_class_1 = classify_image(app_img)
# signature_class_1 = classify_learn.predict(app_img)[1]
# signature_class_1 = str(signature_class_1.item())
if not isinstance(device_img, NoneType):
signature_class_2 = classify_image(device_img)
# signature_class_2 = classify_learn.predict(device_img)[1]
# signature_class_2 = str(signature_class_2.item())
result = ""
similarity_score = -1
if not isinstance(app_img, NoneType) and not isinstance(
device_img, NoneType):
if (config["sig_config"]["similarity_response"][signature_class_1] !=
config["sig_config"]["success_case"] and
config["sig_config"]["similarity_response"][signature_class_2]
!= config["sig_config"]["success_case"]):
result = (
config["sig_config"]["similarity_response"][signature_class_1]
if
config["sig_config"]["similarity_response"][signature_class_1]
!= config["sig_config"]["success_case"] else
config["sig_config"]["similarity_response"][signature_class_2])
if result == "":
featurizer.features = None
similarity_learn.predict(app_img)
app_emb = featurizer.features[0][:]
assert len(app_emb) > 1
featurizer.features = None
similarity_learn.predict(device_img)
device_emb = featurizer.features[0][:]
assert len(device_emb) > 1
featurizer.features = None
similarity_score = vector_distance(app_emb, device_emb)
if similarity_score <= config["sig_config"]["accept_threshold"]:
result = config["sig_config"]["success_case"]
elif similarity_score >= config["sig_config"]["deny_threshold"]:
result = config["sig_config"]["fail_case"]
else:
result = config["sig_config"]["unknown_case"]
elif not isinstance(app_img, NoneType) or not isinstance(
device_img, NoneType):
result = (
app_img
and config["sig_config"]["similarity_response"][signature_class_1]
or config["sig_config"]["similarity_response"][signature_class_2])
else:
return {"status": "error", "description": "no files"}
print("compare signature --- %s seconds ---" % (time.time() - start_time))
return {
"status":
result,
"similarity":
str(similarity_score),
"app_class":
signature_class_1
and config["sig_config"]["classify_response"][signature_class_1] or "",
"device_class":
signature_class_2
and config["sig_config"]["classify_response"][signature_class_2] or "",
}
