def to_masm(self):
return util.postprocess("\n\n".join([
self.directives,
self.enums.to_masm(),
self.structs.to_masm(),
self.segments.to_masm()
]))
def to_masm(self):
return util.postprocess("\n\n".join([
self.directives,
names.get_public_decls(),
self.enums.to_masm(),
self.structs.to_masm(),
self.segments.to_masm()
]))
def test():
CLASSES = []
label_map_path = '../models/labelmap.prototxt'
with open(label_map_path) as f:
lines = f.readlines()
for x in range(3, len(lines), 5):
CLASSES.append(
((lines[x].split(": "))[1]).replace("\"", "").replace("\n", ""))
print(CLASSES)
COLORS = np.random.uniform(0, 255, size=(len(CLASSES), 3))
# frame dimensions should be sqaure
PREPROCESS_DIMS = (300, 300)
DISPLAY_DIMS = (900, 900)
# calculate the multiplier needed to scale the bounding boxes
DISP_MULTIPLIER = DISPLAY_DIMS[0] // PREPROCESS_DIMS[0]
# Load the model
net = cv2.dnn.readNet('../models/no_bn.xml', '../models/no_bn.bin')
# Specify target device
net.setPreferableTarget(cv2.dnn.DNN_TARGET_MYRIAD)
for filename in ["test1.png", "test2.jpg", "test3.jpg"]:
image = cv2.imread(filename)
origimg = np.copy(image)
image = util.preprocess_image(image)
#image = image.transpose((2, 0, 1))
#blob = cv2.dnn.blobFromImage(image, size=PREPROCESS_DIMS)
net.setInput(image)
outputs = net.forward()
box, conf, cls = util.postprocess(origimg, outputs)
for i in range(len(box)):
p1 = (box[i][0], box[i][1])
p2 = (box[i][2], box[i][3])
cv2.rectangle(origimg, p1, p2, (0, 255, 0))
p3 = (max(p1[0], 15), max(p1[1], 15))
title = "%s:%.2f" % (CLASSES[int(cls[i])], conf[i])
cv2.putText(origimg, title, p3, cv2.FONT_ITALIC, 0.6, (0, 255, 0),
1)
cv2.imwrite(filename.split(".")[0] + "SSD.jpg", origimg)
def is_mangled(name, ea):
return util.postprocess(name) != get_demangled_name(ea)
# Load pretrained model
model = models.resnet34(pretrained=True)
# Load and classify image
im = Image.open("cat.jpg") # taken from ImageNet test set
im = im.resize((224, 224), Image.ANTIALIAS)
true_label, true_prob = classify(model, util.preprocess(im))
print("True label: {}, prob: {}".format(true_label, true_prob))
# Generate adversarial example that will correspond to target_class
target_class = 800 #5
print("Target class:", imagenet_labels.label(target_class))
adverserial_image = fgs.fgs(model, util.preprocess(im), target_class,
targeted=True, alpha=0.01, iterations=10,
use_cuda=CUDA)
adverserial_image = util.postprocess(adverserial_image)
adv_label, adv_prob = classify(model, util.preprocess(adverserial_image))
print("Predicted label: {}, prob: {}".format(adv_label, adv_prob))
# Plot results
plt.subplot(131)
plt.title("Before: {} {}%".format(true_label, true_prob))
before_im = np.array(im)
plt.imshow(before_im)
plt.subplot(132)
plt.title("After: {} {}%".format(adv_label, adv_prob))
after_im = np.array(adverserial_image)
plt.imshow(after_im)
plt.subplot(133)
plt.title("Added noise")
noise = after_im - before_im
def is_mangled(name, ea):
return util.postprocess(name) != get_demangled_name(ea)
def to_masm(self):
return util.postprocess("\n\n".join([self.directives,
names.get_public_decls(),
self.enums.to_masm(),
self.structs.to_masm(),
self.segments.to_masm()]))
def to_masm(self):
return util.postprocess("\n\n".join([self.directives,
self.enums.to_masm(),
self.structs.to_masm(),
self.segments.to_masm()]))
best_scale = np.argmax(scales)
new_images = []
for image in images:
new_images.append(
cv.resize(image, scale_list[best_scale]))
feed_dict = {
images_plh: np.array(new_images),
labels_plh: labels,
is_training_plh: True,
best_scale_plh: no_scale
}
pred = sess.run(model.predict, feed_dict=feed_dict)
bboxes, scores, classes = util.postprocess(pred)
map_list.append(
util.compute_map(classes, bboxes, gt_bboxes))
precision_list.append(
util.compute_precision(classes, bboxes,
gt_bboxes))
recall_list.append(
util.compute_recall(classes, bboxes,
gt_bboxes))
if end_of_epoch:
break
test_map = np.mean(map_list)
test_precision = np.mean(precision_list)
