def transform_imglabel(self,image,label,name):
size = np.array(image, dtype=np.uint8).shape
# print(name)
name1 = name
if name1.startswith('site1'):
image = image.resize((math.ceil(2*size[1]),math.ceil(2*size[0])), Image.BICUBIC)
label = label.resize((math.ceil(2*size[1]),math.ceil(2*size[0])), Image.NEAREST)
image = np.asarray(image, np.float32)
print(image.shape)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
elif name1.startswith('site2'):
image = image.resize((math.ceil(2*size[1]),math.ceil(2*size[0])), Image.BICUBIC)
label = label.resize((math.ceil(2*size[1]),math.ceil(2*size[0])), Image.NEAREST)
image = np.asarray(image, np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
elif name1.startswith('site4'):
image = image.resize((math.ceil(1.16*size[1]),math.ceil(1.16*size[0])), Image.BICUBIC)
label = label.resize((math.ceil(1.16*size[1]),math.ceil(1.16*size[0])), Image.NEAREST)
image = np.asarray(image, np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
elif name1.startswith('site3'):
image = np.array(image, dtype=np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
image = np.expand_dims(image, axis=2)
label = np.expand_dims(label, axis=2)
img_label = np.concatenate((image, label), axis=-1)
img_label = CenterCrop(img_label,128)
image = img_label[..., 0]
label = img_label[..., 1]
label = np.squeeze(label)
image = np.squeeze(image)
return image,label
def transform(self, image, label, name):
size = np.array(image, dtype=np.uint8).shape
# print(name)
name1 = name
# print(name)
if name1.startswith('site1') or name1.startswith('site2'):
image = image.resize(
(math.ceil(2 * size[1]), math.ceil(2 * size[0])),
Image.BICUBIC)
label = label.resize(
(math.ceil(2 * size[1]), math.ceil(2 * size[0])),
Image.NEAREST)
image = np.asarray(image, np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
elif name1.startswith('site4'):
image = image.resize(
(math.ceil(1.16 * size[1]), math.ceil(1.16 * size[0])),
Image.BICUBIC)
label = label.resize(
(math.ceil(1.16 * size[1]), math.ceil(1.16 * size[0])),
Image.NEAREST)
image = np.asarray(image, np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
elif name1.startswith('site3'):
image = np.array(image, dtype=np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
label = np.expand_dims(label, axis=2)
img_label = np.concatenate((image, label), axis=-1)
img_label = CenterCrop(img_label, self.img_size)
image = img_label[..., :3]
label = img_label[..., 3]
label = np.squeeze(label)
image = np.squeeze(image)
image = image / 255.0
image = (image - np.mean(image)) / np.std(image)
image = np.transpose(image, (2, 0, 1))
label = np.expand_dims(label, axis=0)
image = torch.FloatTensor(image)
label = torch.FloatTensor(label)
return image, label
def __getitem__(self, index):
datafiles = self.files[index]
image = Image.open(datafiles["img"]).convert('L')
name = datafiles["name"]
label = Image.open(datafiles["lbl"]).convert('L')
if self.set =='test':
if self.resize_and_crop == "center_crop":
size = np.array(label, dtype=np.uint8).shape[1]
if size <= self.crop_size:
image = np.asarray(image, np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
else:
image = np.asarray(image, np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
image = np.expand_dims(image, axis=2)
label = np.expand_dims(label, axis=2)
img_label = np.concatenate((image, label), axis=-1)
img_label = CenterCrop(img_label,self.crop_size)
image = img_label[..., 0]
label = img_label[..., 1]
label = np.squeeze(label)
image = np.squeeze(image)
if self.set =='train':
if self.resize_and_crop == "resize":
image = image.resize(self.img_size, Image.BICUBIC)
image = np.asarray(image, np.float32)
label = label.resize(self.img_size, Image.NEAREST)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
elif self.resize_and_crop == "center_crop":
size = np.array(label, dtype=np.uint8).shape
if size[1] <= self.crop_size:
image = image.resize(self.img_size, Image.BICUBIC)
label = label.resize(self.img_size, Image.NEAREST)
image = np.asarray(image, np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
else:
image = np.asarray(image, np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
image = np.expand_dims(image, axis=2)
label = np.expand_dims(label, axis=2)
img_label = np.concatenate((image, label), axis=-1)
img_label = CenterCrop(img_label,self.crop_size)
image = img_label[..., 0]
label = img_label[..., 1]
label = np.squeeze(label)
image = np.squeeze(image)
elif self.resize_and_crop == "resize_and_center_crop":
image = image.resize(self.img_size, Image.BICUBIC)
image = np.asarray(image, np.float32)
label = label.resize(self.img_size, Image.NEAREST)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
label = np.expand_dims(label, axis=2)
img_label = np.concatenate((image, label), axis=-1)
image, label = CenterCrop(image, label,self.crop_size)
image = img_label[..., 0]
label = img_label[..., 1]
label = np.squeeze(label)
image = np.squeeze(image)
elif self.resize_and_crop == "resize_and_random_crop":
image = image.resize(self.img_size, Image.BICUBIC)
image = np.asarray(image, np.float32)
label = label.resize(self.img_size, Image.NEAREST)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
label = np.expand_dims(label, axis=2)
image = np.expand_dims(image, axis=2)
img_label = np.concatenate((image, label), axis=-1)
img_label = RandomCrop(img_label,self.crop_size)
image = img_label[..., 0]
label = img_label[..., 1]
label = np.squeeze(label)
image = np.squeeze(image)
else:
image = np.asarray(image, np.float32)
label = self.encode_segmap(np.array(label, dtype=np.uint8))
# normalize image
image = image / 255.0
image -= np.mean(image)
image = image/np.std(image)
label = np.expand_dims(label, axis=0)
image = np.expand_dims(image, axis=0)
if self.nlabel:
size = label.shape
lbl = np.zeros((self.n_class,label.shape[1],label.shape[2]))
for i in range(self.n_class):
lbl[i] = np.array((label[0] == i), dtype=np.uint8)
return image.copy(), lbl.copy(), label.copy(), size, name