def __getitem__(self, idx):
print('image path:', self.images[idx])
ori_image = cv2.imread(self.images[idx])
ori_mask = cv2.imread(self.labels[idx], cv2.IMREAD_GRAYSCALE)
train_img, train_mask = crop_resize_data(ori_image, ori_mask)
# Encode
train_mask = encode_labels(train_mask)
sample = [train_img.copy(), train_mask.copy()]
if self.transform:
sample = self.transform(sample)
return sample
def __getitem__(self, item):
# read image data
img_data = cv2.imread(self.image[item])
gray_data = cv2.imread(self.label[item], cv2.IMREAD_GRAYSCALE)
train_img, train_mask = crop_data(img_data, gray_data)
# Encode
train_mask = encode_labels(train_mask)
sample = [train_img, train_mask]
if self.transform:
sample = self.transform(sample)
return sample
def preprocess(self, roi_image, roi_label):
roi_image = roi_image[self.offset:, :]
roi_label = roi_label[self.offset:, :]
train_image = cv2.resize(roi_image,
(self.image_size[0], self.image_size[1]),
interpolation=cv2.INTER_LINEAR)
train_label = cv2.resize(roi_label,
(self.image_size[0], self.image_size[1]),
interpolation=cv2.INTER_NEAREST)
train_label = encode_labels(train_label)
train_image = train_image / (255.0 / 2) - 1
return train_image, train_label
def __getitem__(self, item):
# 在该函数内部对图像做处理,返回一个单个样本
image = cv2.imread(self.images[item], cv2.IMREAD_COLOR)
# print(self.images[item])
label = cv2.imread(self.labels[item], cv2.IMREAD_GRAYSCALE) # 读取灰度图
train_image, train_label = crop_resize_data(image, label)
# 标签encode
train_label = encode_labels(train_label)
sample = [np.copy(train_image), np.copy(train_label)]
# verify_labels(train_label)
if self.transform is not None:
# 需进行数据增强
sample = self.transform(sample)
return sample
def __getitem__(self, idx):
# print('image path:', self.images[idx])
# print('labels path:', self.labels[idx])
try:
ori_image = cv2.imread(self.images[idx])
ori_mask = cv2.imread(self.labels[idx], cv2.IMREAD_GRAYSCALE)
# print ('image shape:', ori_image.shape)
# print('label shape:', ori_mask.shape)
train_img, train_mask = crop_resize_data(ori_image, ori_mask)
# Encode
train_mask = encode_labels(train_mask)
sample = [train_img.copy(), train_mask.copy()]
if self.transform:
sample = self.transform(sample)
return sample
except Exception as e:
print('rechoice id:', idx)
idx = random.choice(range(len(self.images)))
return self.__getitem__(idx)
def train_image_gen(train_list,
batch_size=128,
image_size=(1024, 384),
crop_offset=690):
# 这里要注意的是opencv读取进来的数据格式是BGR格式,因此需要注意的是如何训练的,最后如何预测
# Arrange all indexes
all_batches_index = np.arange(0, len(train_list))
out_images = []
out_masks = []
image_dir = np.array(train_list['Image_Path'])
label_dir = np.array(train_list['Label_Path'])
while True:
# Random shuffle indexes every epoch
np.random.shuffle(all_batches_index)
for index in all_batches_index:
if os.path.exists(image_dir[index]):
image_src = cv2.imread(image_dir[index], cv2.IMREAD_COLOR)
label_src = cv2.imread(label_dir[index], cv2.IMREAD_GRAYSCALE)
roi_image, roi_label = crop_resize_data(
image_src, label_src, image_size, crop_offset)
# Encode
roi_label_encode = encode_labels(roi_label)
out_images.append(roi_image)
out_masks.append(roi_label_encode)
if len(out_images) >= batch_size: # 表示已经满足batch_size的数据量
out_images = np.array(out_images)
out_masks = np.array(out_masks)
# print(out_images.shape)
out_images = out_images[:, :, :, ::-1].transpose(
0, 3, 1, 2).astype(np.float32) / (
255.0 /
2) - 1 # 首先将BGR模式转换为RGB,再进行维度变换,再将数据缩放至[-1,1]
out_masks = out_masks.astype(np.int64)
# print(out_images.shape, out_masks.shape)
yield out_images, out_masks
out_images, out_masks = [], []
else:
print(image_dir, 'does not exist.')
break
def __getitem__(self, item):
label = cv2.imread(self.labels[item], cv2.IMREAD_GRAYSCALE)
train_label = encode_labels(label)
return train_label