def main():
#model = double_stream_model_12()
#model.load_weights("checkpoints/"+"doublestream"+"_model_weights.h5")
from keras.applications.mobilenet import preprocess_input
base_model = MobileNet(weights='imagenet',include_top=False,input_shape=(HEIGHT,WIDTH,3))
preprocessing_function = preprocess_input
class_list_file = "checkpoints/MobileNet_class_list.txt"
class_list = utils.load_class_list(class_list_file)
model = utils.build_finetune_model(base_model,dropout=1e-3,num_classes=len(class_list),fc_layers=[1024,1024])
model.load_weights("checkpoints/MobileNet_model_weights.h5")
path = "/Users/tangxi/Downloads/Compressed/deepfake_baselinev1_1/test_videos"
videos = getvideos(path)
predictor = dlib.shape_predictor("/Users/tangxi/Downloads/Compressed/deepfake_baselinev1_1/shape_predictor_68_face_landmarks.dat")
detector = dlib.get_frontal_face_detector()
predictions=[]
for each in videos:
p_each_video=0.0
vc = cv2.VideoCapture(each)
rval, frame = vc.read()
# 获取视频fps
fps = vc.get(cv2.CAP_PROP_FPS)
# 获取视频总帧数
frame_all = vc.get(cv2.CAP_PROP_FRAME_COUNT)
print("[INFO] 视频FPS: {}".format(fps))
print("[INFO] 视频总帧数: {}".format(frame_all))
print("[INFO] 视频时长: {}s".format(frame_all/fps))
fake_count=0
total_count=0
while True:
ret,frame = vc.read()
if ret is False:
break
total_count+=1
img = frame.copy()
gray = cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY)
dets = detector(gray, 0)
if len(dets)!=1:continue
d = dets[0]
x1 = d.top() if d.top() > 0 else 0
y1 = d.bottom() if d.bottom() > 0 else 0
x2 = d.left() if d.left() > 0 else 0
y2 = d.right() if d.right() > 0 else 0
face = img[x1-25:y1+10,x2-8:y2+8]
print(face.shape)
p_fake = predict_single_image(model,face)
if p_fake>0.5:
fake_count+=1
if fake_count/float(total_count)>0.5:
p_each_video = fake_count/float(total_count)
else:
p_each_video = 0.5
predictions.append(p_each_video)
submission_df = pd.DataFrame({"filename": test_videos, "label": predictions})
submission_df.to_csv("submission.csv", index=False)
def build_non_bottleneck_top_model(base_model, class_list):
finetune_model = utils.build_finetune_model(base_model,
dropout=args.dropout,
fc_layers=FC_LAYERS,
num_classes=len(class_list))
if args.continue_training:
finetune_model.load_weights(WEIGHTS_PATH)
adam = Adam(lr=0.00001)
finetune_model.compile(adam,
loss='categorical_crossentropy',
metrics=['accuracy'])
return finetune_model
preprocessing_function=preprocessing_function)
train_generator = train_datagen.flow_from_directory(TRAIN_DIR,
target_size=(HEIGHT,
WIDTH),
batch_size=BATCH_SIZE)
validation_generator = val_datagen.flow_from_directory(
VAL_DIR, target_size=(HEIGHT, WIDTH), batch_size=BATCH_SIZE)
# Save the list of classes for prediction mode later
class_list = utils.get_subfolders(TRAIN_DIR)
utils.save_class_list(class_list, model_name=args.model, dataset_name="")
finetune_model = utils.build_finetune_model(base_model,
dropout=args.dropout,
fc_layers=FC_LAYERS,
num_classes=len(class_list))
if args.continue_training:
finetune_model.load_weights("./checkpoints/" + args.model +
"_model_weights.h5")
print("load success!")
adam = Adam(lr=0.00001)
finetune_model.compile(adam,
loss='categorical_crossentropy',
metrics=['accuracy'])
num_train_images = utils.get_num_files(TRAIN_DIR)
num_val_images = utils.get_num_files(VAL_DIR)
base_model = MobileNet(weights='imagenet',
include_top=False,
input_shape=(HEIGHT, WIDTH, 3))
elif model == "ResNet50":
HEIGHT = 224
WIDTH = 224
from keras.applications.resnet50 import preprocess_input
preprocessing_function = preprocess_input
base_model = ResNet50(weights='imagenet',
include_top=False,
input_shape=(HEIGHT, WIDTH, 3))
class_list_file = "./class_list.txt"
class_list = utils.load_class_list(class_list_file)
finetune_model = utils.build_finetune_model(base_model,
dropout=DROPOUT,
fc_layers=FC_LAYERS,
num_classes=len(class_list))
finetune_model.load_weights("./" + model + "_model_weights.h5")
def classify(image):
global finetune_model
try:
image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
except:
print("ERROR classify: could not convert image into color",
image.shape)
return 0
try:
image = np.float32(cv2.resize(
class_list,
model_name=args.model,
dataset_name=os.path.basename(args.dataset),
)
optim = eval(args.optimizer)(lr=args.lr)
if args.continue_training is not None:
finetune_model = load_model(args.continue_training)
if args.transfer_strategy == "finetune":
utils.set_trainable(finetune_model, True)
else:
finetune_model = utils.build_finetune_model(
base_model,
dropout=args.dropout,
fc_layers=FC_LAYERS,
num_classes=len(class_list),
as_fixed_feature_extractor=True
if args.transfer_strategy == "fixed"
else False,
skip_interval=args.skip_interval,
)
finetune_model.compile(optim, loss="categorical_crossentropy", metrics=["accuracy"])
if args.summarize_model:
finetune_model.summary()
num_train_images = utils.get_num_files(TRAIN_DIR)
num_val_images = utils.get_num_files(VAL_DIR)
def lr_decay(epoch):
if epoch % 20 == 0 and epoch != 0:
lr = K.get_value(model.optimizer.lr)