def extract_faces(data_dir):
face_detector = MTCNN(device=device, margin=16)
face_detector.eval()
for image_name in tqdm(os.listdir(data_dir), desc='Extracting faces'):
inp_img_path = os.path.join(data_dir, image_name)
out_img_path = os.path.join(FACES_DATA_DIR, image_name)
if not os.path.exists(out_img_path):
image = Image.open(inp_img_path)
face_detector(image, save_path=out_img_path)
layers=[3, 4, 6, 3],
groups=32,
width_per_group=4)
# Override the existing FC layer with a new one.
self.fc = nn.Linear(2048, num_classes)
if checkpoint:
self.load_state_dict(checkpoint)
# load model - ResNext
checkpoint_path = "runs/May12_02-47-20_4497d0b803cd/epc_12.pth" #"models/Apr07_17-35-04_bb90ec5a54cc/model.pt"
net = MyResNeXt(torch.load(checkpoint_path))
net.eval()
margin = 16
mtcnn = MTCNN(device='cuda', margin=margin)
mtcnn.eval()
# get data
test_imgs_folder = "../dataset/faceforensics_benchmark_images"
test_imgs = glob(test_imgs_folder + "/*.png")
result_dict = dict()
transform = transforms.Compose([
transforms.Resize((160, 160)),
transforms.ToTensor(),
transforms.Normalize((123.675, 116.28, 103.53), (58.395, 57.12, 57.375))
])
with torch.no_grad():
for test_img in tqdm(test_imgs):
img = Image.open(test_img)
