def transform(self, pil, mode="sketch"):
def show(mode, pil_numpy):
print(mode, ",".join([str(i) for i in pil_numpy.flatten() if i != 0]))
pil_numpy = np.array(pil)
if len(pil_numpy.shape) == 2:
if self.edge_map and mode == "image":
pil_numpy = cv2.Canny(pil_numpy, 100, 200)
#show("edge",pil_numpy)
pil_numpy = to_rgb(pil_numpy)
#pil_numpy = np.tile(pil_numpy,3).reshape(pil_numpy.shape+(-1,))
elif pil_numpy.shape[2] == 4:
#show("sketch",pil_numpy[:,:,3])
pil_numpy = to_rgb(pil_numpy[:,:,3])
#pil_numpy = np.tile(pil_numpy[:,:,3],3).reshape(pil_numpy.shape[0:2]+(-1,))
#pil_numpy[:,:,2] = 0
if self.opt.image_type == 'EDGE':
gray_pil = Image.fromarray(pil_numpy)
pil_numpy = np.array(gray_pil.convert('L'))
pil_numpy = to_rgb(pil_numpy)
pil_numpy = cv2.resize(pil_numpy,(self.opt.scale_size,self.opt.scale_size))
if self.transform_fun is not None:
pil_numpy = self.transform_fun(pil_numpy)
#data_info.write(",".join([str(i) for i in pil_numpy.numpy().flatten() if i != 0])+"\n")
return pil_numpy
def transform(self, pil):
pil = pil.convert('RGB')
pil_numpy = np.array(pil)
if len(pil_numpy.shape) == 2:
pil_numpy = to_rgb(pil_numpy)
elif pil_numpy.shape[2] == 4:
pil_numpy = to_rgb(pil_numpy[:,:,3])
pil_numpy = cv2.resize(pil_numpy,(self.opt.scale_size, self.opt.scale_size))
if self.transform_fun is not None:
pil = Image.fromarray(pil_numpy)
pil_numpy = self.transform_fun(pil)
return pil_numpy
def load_image(self, pil):
def show(mode, pil_numpy):
print(
mode,
len(",".join([str(i) for i in pil_numpy.flatten() if i != 0])))
if self.opt.image_type == 'RGB':
pil = pil.convert('RGB')
pil_numpy = np.array(pil)
elif self.opt.image_type == 'GRAY':
pil = pil.convert('L')
pil_numpy = np.array(pil)
elif self.opt.image_type == 'EDGE':
pil = pil.convert('L')
pil_numpy = np.array(pil)
#show('edge', pil_numpy)
pil_numpy = cv2.Canny(pil_numpy, 0, 200)
if 'densenet' in self.opt.feature_model and not self.opt.image_type == 'RGB':
pil_numpy = to_rgb(pil_numpy)
#print('image{}'.format(pil_numpy.shape))
#if self.opt.image_type == 'GRAY' or self.opt.image_type == 'EDGE':
# pil_numpy = pil_numpy.reshape(pil_numpy.shape + (1,))
#pil_numpy = cv2.resize(pil_numpy, (self.opt.scale_size, self.opt.scale_size))
#if self.opt.sketch_type == 'GRAY' or self.opt.image_type == 'EDGE':
# pil_numpy = pil_numpy.reshape(pil_numpy.shape[:2])
transform_fun = self.transform_fun if self.mode == 'train' else self.test_transform_fun
if transform_fun is not None:
pil = Image.fromarray(pil_numpy)
pil_numpy = transform_fun(pil)
return pil_numpy
def transform(self, pil):
pil = pil.convert('RGB')
pil_numpy = np.array(pil)
if len(pil_numpy.shape) == 2:
pil_numpy = to_rgb(pil_numpy)
#pil_numpy = np.tile(pil_numpy,3).reshape(pil_numpy.shape+(-1,))
elif pil_numpy.shape[2] == 4:
pil_numpy = to_rgb(pil_numpy[:, :, 3])
#pil_numpy = np.tile(pil_numpy[:,:,3],3).reshape(pil_numpy.shape[0:2]+(-1,))
#pil_numpy[:,:,2] = 0
pil_numpy = cv2.resize(pil_numpy,
(self.opt.scale_size, self.opt.scale_size))
if self.transform_fun is not None:
pil = Image.fromarray(pil_numpy)
pil_numpy = self.transform_fun(pil)
#data_info.write(",".join([str(i) for i in pil_numpy.numpy().flatten() if i != 0])+"\n")
return pil_numpy
def load_sketch(self, pil):
def show(mode, pil_numpy):
print(mode, len(",".join([str(i) for i in pil_numpy.flatten() if i != 0])))
pil = pil.convert('L')
pil_numpy = np.array(pil)
if self.opt.sketch_type == 'RGB' or 'densenet' in self.opt.feature_model:
pil_numpy = to_rgb(pil_numpy)
transform_fun = self.transform_fun if self.mode == 'train' else self.test_transform_fun
if transform_fun is not None:
pil = Image.fromarray(pil_numpy)
pil_numpy = transform_fun(pil)
return pil_numpy
def load_image(self, pil):
def show(mode, pil_numpy):
print(mode, len(",".join([str(i) for i in pil_numpy.flatten() if i != 0])))
if self.opt.image_type == 'RGB':
pil = pil.convert('RGB')
pil_numpy = np.array(pil)
elif self.opt.image_type == 'GRAY':
pil = pil.convert('L')
pil_numpy = np.array(pil)
elif self.opt.image_type == 'EDGE':
pil = pil.convert('L')
pil_numpy = np.array(pil)
#show('edge', pil_numpy)
pil_numpy = cv2.Canny(pil_numpy, 0, 200)
if 'densenet' in self.opt.feature_model and not self.opt.image_type == 'RGB':
pil_numpy = to_rgb(pil_numpy)
transform_fun = self.transform_fun if self.mode == 'train' else self.test_transform_fun
if transform_fun is not None :
pil = Image.fromarray(pil_numpy)
pil_numpy = transform_fun(pil)
return pil_numpy