def generate(self, test=False):
while True:
batch_X = []
batch_Y = []
cur_batch_path = np.random.choice(self.all_id, 32)
for p in cur_batch_path:
cur_y = self.id2label[p]
if test and p == -1:
continue
# protect class images in train dataset
elif p == -1:
cur_x = random.choice(self.protect_images_train)
else:
if test:
cur_path = random.choice(self.id2pathtest[p])
else:
cur_path = random.choice(self.id2path[p])
im = image.load_img(cur_path, target_size=(224, 224))
cur_x = image.img_to_array(im)
cur_x = preprocess(cur_x, 'imagenet')
batch_X.append(cur_x)
batch_Y.append(cur_y)
batch_X = np.array(batch_X)
batch_Y = to_categorical(np.array(batch_Y),
num_classes=self.num_classes)
yield batch_X, batch_Y
def clipping(self, imgs):
imgs = reverse_preprocess(imgs, self.intensity_range)
imgs = np.clip(imgs, 0, self.max_val)
imgs = preprocess(imgs, self.intensity_range)
return imgs