def __init__(self, args):
super(NeuralStyleTransfer, self).__init__(args=args)
self._style_loss = loss.VGGStyleLoss(transfer_mode=args.transfer_mode,
resize=True)
npzfile = np.load(args.vector_file)
self.x_ctt = torch.tensor(npzfile['x_ctt']).to(device)
self.x_color = torch.tensor(npzfile['x_color']).to(device)
self.x_alpha = torch.tensor(npzfile['x_alpha']).to(device)
self.m_grid = int(np.sqrt(self.x_ctt.shape[0]))
self.anchor_id = self.x_ctt.shape[1] - 1
img_ = cv2.imread(args.content_img_path, cv2.IMREAD_COLOR)
img_ = cv2.cvtColor(img_, cv2.COLOR_BGR2RGB).astype(np.float32) / 255.
self.img_ = cv2.resize(img_, (128 * self.m_grid, 128 * self.m_grid))
self.img_batch = utils.img2patches(self.img_, self.m_grid).to(device)
style_img = cv2.imread(args.style_img_path, cv2.IMREAD_COLOR)
self.style_img_ = cv2.cvtColor(style_img, cv2.COLOR_BGR2RGB).astype(
np.float32) / 255.
self.style_img = cv2.blur(cv2.resize(self.style_img_, (128, 128)),
(2, 2))
self.style_img = torch.tensor(self.style_img).permute(
[2, 0, 1]).unsqueeze(0).to(device)
def __init__(self, args):
super(NeuralStyleTransfer, self).__init__(args=args)
self._style_loss = loss.VGGStyleLoss(transfer_mode=args.transfer_mode, resize=True)
print('loading pre-generated vector file...')
if os.path.exists(args.vector_file) is False:
exit('vector file does not exist, pls check --vector_file, or run demo.py fist')
else:
npzfile = np.load(args.vector_file)
self.x_ctt = torch.tensor(npzfile['x_ctt']).to(device)
self.x_color = torch.tensor(npzfile['x_color']).to(device)
self.x_alpha = torch.tensor(npzfile['x_alpha']).to(device)
self.m_grid = int(np.sqrt(self.x_ctt.shape[0]))
self.anchor_id = self.x_ctt.shape[1] - 1
img_ = cv2.imread(args.content_img_path, cv2.IMREAD_COLOR)
img_ = cv2.cvtColor(img_, cv2.COLOR_BGR2RGB).astype(np.float32) / 255.
self.img_ = cv2.resize(img_, (128*self.m_grid, 128*self.m_grid))
self.img_batch = utils.img2patches(self.img_, self.m_grid).to(device)
style_img = cv2.imread(args.style_img_path, cv2.IMREAD_COLOR)
self.style_img_ = cv2.cvtColor(style_img, cv2.COLOR_BGR2RGB).astype(np.float32) / 255.
self.style_img = cv2.blur(cv2.resize(self.style_img_, (128, 128)), (2, 2))
self.style_img = torch.tensor(self.style_img).permute([2, 0, 1]).unsqueeze(0).to(device)
def __init__(self, args):
super(Painter, self).__init__(args=args)
self.m_grid = args.m_grid
self.max_m_strokes = args.max_m_strokes
self.img_path = args.img_path
self.img_ = cv2.imread(args.img_path, cv2.IMREAD_COLOR)
self.img_ = (
cv2.cvtColor(self.img_, cv2.COLOR_BGR2RGB).astype(np.float32) /
255.0)
self.input_aspect_ratio = self.img_.shape[0] / self.img_.shape[1]
self.img_ = cv2.resize(
self.img_,
(self.net_G.out_size * args.m_grid,
self.net_G.out_size * args.m_grid),
cv2.INTER_AREA,
)
self.m_strokes_per_block = int(args.max_m_strokes /
(args.m_grid * args.m_grid))
self.img_batch = utils.img2patches(self.img_, args.m_grid,
self.net_G.out_size).to(device)
self.final_rendered_images = None
def __init__(self, args):
super(Painter, self).__init__(args=args)
self.m_grid = args.m_grid
self.max_m_strokes = args.max_m_strokes
self.img_path = args.img_path
self.img_ = cv2.imread(args.img_path, cv2.IMREAD_COLOR)
self.img_ = cv2.cvtColor(self.img_, cv2.COLOR_BGR2RGB).astype(np.float32) / 255.
self.img_ = cv2.resize(self.img_, (128 * args.m_grid, 128 * args.m_grid))
self.m_strokes_per_block = int(args.max_m_strokes / (args.m_grid * args.m_grid))
self.img_batch = utils.img2patches(self.img_, args.m_grid).to(device)
self.final_rendered_images = None
