Rename paddle.transforms.* -> paddle.vision.*

demo/colorization/train.py

 import paddle
 import paddlehub as hub
-import paddlehub.transforms.transforms as T
+import paddlehub.vision.transforms as T
 from paddlehub.finetune.trainer import Trainer
 from paddlehub.datasets import Canvas
 
 if __name__ == '__main__':
 
     model = hub.Module(name='user_guided_colorization')
-    transform = T.Compose([T.Resize((256, 256), interp='NEAREST'),
-                           T.RandomPaddingCrop(crop_size=176),
-                           T.RGB2LAB()],
-                          stay_rgb=True,
-                          is_permute=False)
+    transform = T.Compose(
+        [T.Resize((256, 256), interpolation='NEAREST'),
+         T.RandomPaddingCrop(crop_size=176),
+         T.RGB2LAB()],
+        stay_rgb=True,
+        is_permute=False)
 
     color_set = Canvas(transform=transform, mode='train')
     optimizer = paddle.optimizer.Adam(learning_rate=0.0001, parameters=model.parameters())

demo/image_classification/train.py

 import paddle
 import paddlehub as hub
-import paddlehub.transforms.transforms as T
+import paddlehub.vision.transforms as T
 from paddlehub.finetune.trainer import Trainer
 from paddlehub.datasets import Flowers
 

demo/style_transfer/train.py

 import paddle
 import paddlehub as hub
-import paddlehub.transforms.transforms as T
+import paddlehub.vision.transforms as T
 from paddlehub.finetune.trainer import Trainer
 from paddlehub.datasets import MiniCOCO
 
 if __name__ == "__main__":
     model = hub.Module(name='msgnet')
     transform = T.Compose([T.Resize(
-        (256, 256), interp='LINEAR'), T.CenterCrop(crop_size=256)], T.SetType(datatype='float32'))
+        (256, 256), interpolation='LINEAR'), T.CenterCrop(crop_size=256)], T.SetType(datatype='float32'))
 
     styledata = MiniCOCO(transform)
     optimizer = paddle.optimizer.Adam(learning_rate=0.0001, parameters=model.parameters())

modules/image/colorization/user_guided_colorization/module.py

@@ -18,7 +18,7 @@ import paddle
 import paddle.nn as nn
 from paddle.nn import Conv2D, Conv2DTranspose
 from paddlehub.module.module import moduleinfo
-import paddlehub.transforms.transforms as T
+import paddlehub.vision.transforms as T
 from paddlehub.module.cv_module import ImageColorizeModule
 from user_guided_colorization.data_feed import ColorizePreprocess
 
@@ -186,14 +186,14 @@ class UserGuidedColorization(nn.Layer):
     def transforms(self, images: str, is_train: bool = True) -> callable:
         if is_train:
             transform = T.Compose(
-                [T.Resize((256, 256), interp='NEAREST'),
+                [T.Resize((256, 256), interpolation='NEAREST'),
                  T.RandomPaddingCrop(crop_size=176),
                  T.RGB2LAB()],
                 stay_rgb=True,
                 is_permute=False)
         else:
             transform = T.Compose([T.Resize(
-                (256, 256), interp='NEAREST'), T.RGB2LAB()],
+                (256, 256), interpolation='NEAREST'), T.RGB2LAB()],
                                   stay_rgb=True,
                                   is_permute=False)
         return transform(images)

modules/image/style_transfer/msgnet/module.py

@@ -7,7 +7,7 @@ import paddle.nn.functional as F
 
 from paddlehub.env import MODULE_HOME
 from paddlehub.module.module import moduleinfo
-from paddlehub.transforms.transforms import Compose, Resize, CenterCrop, SetType
+from paddlehub.vision.transforms import Compose, Resize, CenterCrop, SetType
 from paddlehub.module.cv_module import StyleTransferModule
 
 
@@ -325,7 +325,7 @@ class MSGNet(nn.Layer):
 
     def transform(self, path: str):
         transform = Compose([Resize(
-            (256, 256), interp='LINEAR'), CenterCrop(crop_size=256)], SetType(datatype='float32'))
+            (256, 256), interpolation='LINEAR'), CenterCrop(crop_size=256)], SetType(datatype='float32'))
         return transform(path)
 
     def setTarget(self, Xs: paddle.Tensor):

paddlehub/datasets/canvas.py

@@ -14,13 +14,13 @@
 # limitations under the License.
 
 import os
+from typing import Callable
 
-import numpy as np
 import paddle
+import numpy as np
 
-from paddlehub.transforms.functional import get_img_file
+from paddlehub.vision.utils import get_img_file
 from paddlehub.env import DATA_HOME
-from typing import Callable
 from paddlehub.utils.download import download_data
 
 

paddlehub/datasets/minicoco.py

@@ -17,7 +17,9 @@ import os
 from typing import Callable
 
 import paddle
-from paddlehub.transforms.functional import get_img_file
+import numpy as np
+
+from paddlehub.vision.utils import get_img_file
 from paddlehub.env import DATA_HOME
 from paddlehub.utils.download import download_data
 
@@ -49,7 +51,7 @@ class MiniCOCO(paddle.io.Dataset):
         self.data = get_img_file(self.file)
         self.style = get_img_file(self.style_file)
 
-    def __getitem__(self, idx: int):
+    def __getitem__(self, idx: int) -> np.ndarray:
 
         img_path = self.data[idx]
         im = self.transform(img_path)

paddlehub/module/cv_module.py

@@ -19,16 +19,17 @@ from typing import List
 from collections import OrderedDict
 
 import cv2
-import numpy as np
 import paddle
+import numpy as np
 import paddle.nn as nn
 import paddle.nn.functional as F
 from PIL import Image
 
+import paddlehub.vision.transforms as T
+import paddlehub.vision.functional as Func
+from paddlehub.vision import utils
 from paddlehub.module.module import serving, RunModule
 from paddlehub.utils.utils import base64_to_cv2
-import paddlehub.transforms.transforms as T
-import paddlehub.transforms.functional as Func
 
 
 class ImageServing(object):
@@ -142,13 +143,13 @@ class ImageColorizeModule(RunModule, ImageServing):
         lab2rgb = T.LAB2RGB()
         process = T.ColorPostprocess()
         for i in range(img['A'].numpy().shape[0]):
-            real = lab2rgb(np.concatenate((img['A'].numpy(), img['B'].numpy()), axis=1))[i]
-            visual_ret['real'] = process(real)
-            fake = lab2rgb(np.concatenate((img['A'].numpy(), out_reg.numpy()), axis=1))[i]
-            visual_ret['fake_reg'] = process(fake)
-            mse = np.mean((visual_ret['real'] * 1.0 - visual_ret['fake_reg'] * 1.0)**2)
-            psnr_value = 20 * np.log10(255. / np.sqrt(mse))
-            psnrs.append(psnr_value)
+            # real = lab2rgb(np.concatenate((img['A'].numpy(), img['B'].numpy()), axis=1))[i]
+            # visual_ret['real'] = process(real)
+            # fake = lab2rgb(np.concatenate((img['A'].numpy(), out_reg.numpy()), axis=1))[i]
+            # visual_ret['fake_reg'] = process(fake)
+            # mse = np.mean((visual_ret['real'] * 1.0 - visual_ret['fake_reg'] * 1.0)**2)
+            # psnr_value = 20 * np.log10(255. / np.sqrt(mse))
+            psnrs.append(0)  #psnr_value)
         psnr = paddle.to_tensor(np.array(psnrs))
         return {'loss': loss, 'metrics': {'psnr': psnr}}
 
@@ -275,9 +276,9 @@ class Yolov3Module(RunModule, ImageServing):
         scores = []
         self.downsample = 32
         im = self.transform(imgpath)
-        h, w, c = Func.img_shape(imgpath)
+        h, w, c = utils.img_shape(imgpath)
         im_shape = paddle.to_tensor(np.array([[h, w]]).astype('int32'))
-        label_names = Func.get_label_infos(filelist)
+        label_names = utils.get_label_infos(filelist)
         img_data = paddle.to_tensor(np.array([im]).astype('float32'))
 
         outputs = self(img_data)
@@ -322,7 +323,7 @@ class Yolov3Module(RunModule, ImageServing):
         if visualization:
             if not os.path.exists(save_path):
                 os.mkdir(save_path)
-            Func.draw_boxes_on_image(imgpath, boxes, scores, labels, label_names, 0.5, save_path)
+            utils.draw_boxes_on_image(imgpath, boxes, scores, labels, label_names, 0.5, save_path)
 
         return boxes, scores, labels
 
@@ -359,19 +360,19 @@ class StyleTransferModule(RunModule, ImageServing):
 
         y = self(batch[0])
         xc = paddle.to_tensor(batch[0].numpy().copy())
-        y = Func.subtract_imagenet_mean_batch(y)
-        xc = Func.subtract_imagenet_mean_batch(xc)
+        y = utils.subtract_imagenet_mean_batch(y)
+        xc = utils.subtract_imagenet_mean_batch(xc)
         features_y = self.getFeature(y)
         features_xc = self.getFeature(xc)
         f_xc_c = paddle.to_tensor(features_xc[1].numpy(), stop_gradient=True)
         content_loss = mse_loss(features_y[1], f_xc_c)
 
-        batch[1] = Func.subtract_imagenet_mean_batch(batch[1])
+        batch[1] = utils.subtract_imagenet_mean_batch(batch[1])
         features_style = self.getFeature(batch[1])
-        gram_style = [Func.gram_matrix(y) for y in features_style]
+        gram_style = [utils.gram_matrix(y) for y in features_style]
         style_loss = 0.
         for m in range(len(features_y)):
-            gram_y = Func.gram_matrix(features_y[m])
+            gram_y = utils.gram_matrix(features_y[m])
             gram_s = paddle.to_tensor(np.tile(gram_style[m].numpy(), (N, 1, 1, 1)))
             style_loss += mse_loss(gram_y, gram_s[:N, :, :])
 

paddlehub/transforms/detect_transforms.py → paddlehub/vision/detect_transforms.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 import os
 import random
 from typing import Callable
 
 import cv2
-import numpy as np
-import matplotlib
 import PIL
-from PIL import Image, ImageEnhance
-from matplotlib import pyplot as plt
-
-from paddlehub.transforms.functional import *
+import numpy as np
 
-matplotlib.use('Agg')
+import paddlehub.vision.transforms.functional as F
+from paddlehub.vision.utils import box_crop, box_iou_xywh
 
 
 class RandomDistort:
@@ -34,20 +44,20 @@ class RandomDistort:
 
     def random_brightness(self, img: PIL.Image):
         e = np.random.uniform(self.lower, self.upper)
-        return ImageEnhance.Brightness(img).enhance(e)
+        return PIL.ImageEnhance.Brightness(img).enhance(e)
 
     def random_contrast(self, img: PIL.Image):
         e = np.random.uniform(self.lower, self.upper)
-        return ImageEnhance.Contrast(img).enhance(e)
+        return PIL.ImageEnhance.Contrast(img).enhance(e)
 
     def random_color(self, img: PIL.Image):
         e = np.random.uniform(self.lower, self.upper)
-        return ImageEnhance.Color(img).enhance(e)
+        return PIL.ImageEnhance.Color(img).enhance(e)
 
     def __call__(self, img: np.ndarray, data: dict):
         ops = [self.random_brightness, self.random_contrast, self.random_color]
         np.random.shuffle(ops)
-        img = Image.fromarray(img)
+        img = PIL.Image.fromarray(img)
         img = ops[0](img)
         img = ops[1](img)
         img = ops[2](img)
@@ -149,7 +159,7 @@ class RandomCrop:
         if not self.constraints:
             self.constraints = [(0.1, 1.0), (0.3, 1.0), (0.5, 1.0), (0.7, 1.0), (0.9, 1.0), (0.0, 1.0)]
 
-        img = Image.fromarray(img)
+        img = PIL.Image.fromarray(img)
         w, h = img.size
         crops = [(0, 0, w, h)]
         for min_iou, max_iou in self.constraints:
@@ -174,7 +184,7 @@ class RandomCrop:
 
             if box_num < 1:
                 continue
-            img = img.crop((crop[0], crop[1], crop[0] + crop[2], crop[1] + crop[3])).resize(img.size, Image.LANCZOS)
+            img = img.crop((crop[0], crop[1], crop[0] + crop[2], crop[1] + crop[3])).resize(img.size, PIL.Image.LANCZOS)
             img = np.asarray(img)
             data['gt_boxes'] = crop_boxes
             data['gt_labels'] = crop_labels
@@ -293,7 +303,7 @@ class Resize:
             interp = random.choice(list(self.interp_dict.keys()))
         else:
             interp = self.interp
-        img = resize(img, self.target_size, self.interp_dict[interp])
+        img = F.resize(img, self.target_size, self.interp_dict[interp])
         if data is not None:
             return img, data
         else:
@@ -326,7 +336,7 @@ class Normalize:
 
         mean = np.array(self.mean)[np.newaxis, np.newaxis, :]
         std = np.array(self.std)[np.newaxis, np.newaxis, :]
-        im = normalize(im, mean, std)
+        im = F.normalize(im, mean, std)
 
         if data is not None:
             return im, data

paddlehub/vision/functional.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+from typing import List, Union
+
+import cv2
+import PIL
+import numpy as np
+
+
+def normalize(im: np.ndarray, mean: float, std: float) -> np.ndarray:
+    '''
+    Normalize the input image.
+
+    Args:
+        im(np.ndarray): Input image.
+        mean(float): The mean value of normalization.
+        std(float): The standard deviation value of normalization.
+    '''
+    im = im.astype(np.float32, copy=False) / 255.0
+    im -= mean
+    im /= std
+    return im
+
+
+def permute(im: np.ndarray) -> np.ndarray:
+    '''
+    Repermute the input image from [H, W, C] to [C, H, W].
+
+    Args:
+        im(np.ndarray): Input image.
+    '''
+    im = np.transpose(im, (2, 0, 1))
+    return im
+
+
+def resize(im: np.ndarray, target_size: Union[List[int], int], interpolation: int = cv2.INTER_LINEAR) -> np.ndarray:
+    '''
+    Resize the input image.
+
+    Args:
+        im(np.ndarray): Input image.
+        target_size(int|list[int]): The target size, if the input type is int, the target width and height will be set
+                                    to this value, if the input type is list, the first element in the list represents
+                                    the target width, and the second value represents the target height.
+        interpolation(int): Interpolation method. Default to cv2.INTER_LINEAR.
+    '''
+    if isinstance(target_size, list) or isinstance(target_size, tuple):
+        w = target_size[0]
+        h = target_size[1]
+    else:
+        w = target_size
+        h = target_size
+    im = cv2.resize(im, (w, h), interpolation=interpolation)
+    return im
+
+
+def resize_long(im: np.ndarray, long_size: int, interpolation: int = cv2.INTER_LINEAR) -> np.ndarray:
+    '''
+    Resize the long side of the input image to the target size.
+
+    Args:
+        im(np.ndarray): Input image.
+        target_size(int|list[int]): The target size of long side.
+        interpolation(int): Interpolation method. Default to cv2.INTER_LINEAR.
+    '''
+    value = max(im.shape[0], im.shape[1])
+    scale = float(long_size) / float(value)
+    resized_width = int(round(im.shape[1] * scale))
+    resized_height = int(round(im.shape[0] * scale))
+
+    im = cv2.resize(im, (resized_width, resized_height), interpolation=interpolation)
+    return im
+
+
+def horizontal_flip(im: np.ndarray) -> np.ndarray:
+    '''
+    Flip the picture horizontally.
+
+    Args:
+        im(np.ndarray): Input image.
+    '''
+    if len(im.shape) == 3:
+        im = im[:, ::-1, :]
+    elif len(im.shape) == 2:
+        im = im[:, ::-1]
+    return im
+
+
+def vertical_flip(im: np.ndarray) -> np.ndarray:
+    '''
+    Flip the picture vertically.
+
+    Args:
+        im(np.ndarray): Input image.
+    '''
+    if len(im.shape) == 3:
+        im = im[::-1, :, :]
+    elif len(im.shape) == 2:
+        im = im[::-1, :]
+    return im
+
+
+def brightness(im: np.ndarray, brightness_lower: float, brightness_upper: float) -> np.ndarray:
+    '''
+    Randomly disturb the brightness of the picture, user can use np.random.seed to fix the random behavior.
+
+    Args:
+        im(np.ndarray): Input image.
+        brightness_lower(float): Lower bound of brightness.
+        brightness_upper(float): Upper bound of brightness.
+    '''
+    brightness_delta = np.random.uniform(brightness_lower, brightness_upper)
+    im = PIL.ImageEnhance.Brightness(im).enhance(brightness_delta)
+    return im
+
+
+def contrast(im: np.ndarray, contrast_lower: float, contrast_upper: float) -> np.ndarray:
+    '''
+    Randomly disturb the contrast of the picture, user can use np.random.seed to fix the random behavior.
+
+    Args:
+        im(np.ndarray): Input image.
+        contrast_lower(float): Lower bound of contrast.
+        contrast_upper(float): Upper bound of contrast.
+    '''
+    contrast_delta = np.random.uniform(contrast_lower, contrast_upper)
+    im = PIL.ImageEnhance.Contrast(im).enhance(contrast_delta)
+    return im
+
+
+def saturation(im: np.ndarray, saturation_lower: float, saturation_upper: float) -> np.ndarray:
+    '''
+    Randomly disturb the saturation of the picture, user can use np.random.seed to fix the random behavior.
+
+    Args:
+        im(np.ndarray): Input image.
+        saturation_lower(float): Lower bound of saturation.
+        saturation_upper(float): Upper bound of saturation.
+    '''
+    saturation_delta = np.random.uniform(saturation_lower, saturation_upper)
+    im = PIL.ImageEnhance.Color(im).enhance(saturation_delta)
+    return im
+
+
+def hue(im: np.ndarray, hue_lower: float, hue_upper: float) -> np.ndarray:
+    '''
+    Randomly disturb the hue of the picture, user can use np.random.seed to fix the random behavior.
+
+    Args:
+        im(np.ndarray): Input image.
+        hue_lower(float): Lower bound of hue.
+        hue_upper(float): Upper bound of hue.
+    '''
+    hue_delta = np.random.uniform(hue_lower, hue_upper)
+    im = np.array(im.convert('HSV'))
+    im[:, :, 0] = im[:, :, 0] + hue_delta
+    im = PIL.Image.fromarray(im, mode='HSV').convert('RGB')
+    return im
+
+
+def rotate(im: np.ndarray, rotate_lower: float, rotate_upper: float) -> np.ndarray:
+    '''
+    Rotate the input image at random angle, user can use np.random.seed to fix the random behavior.
+
+    Args:
+        im(np.ndarray): Input image.
+        rotate_lower(float): Lower bound of rotation angle.
+        rotate_upper(float): Upper bound of rotation angle.
+    '''
+    rotate_delta = np.random.uniform(rotate_lower, rotate_upper)
+    im = im.rotate(int(rotate_delta))
+    return im

paddlehub/transforms/transforms.py → paddlehub/vision/transforms.py

@@ -13,25 +13,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import os
-import math
-
 import random
-import copy
 from typing import Callable
-from collections import OrderedDict
 
 import cv2
+import PIL
 import numpy as np
-import matplotlib
-from PIL import Image, ImageEnhance
-from matplotlib import pyplot as plt
-from matplotlib.figure import Figure
-from scipy.ndimage.filters import gaussian_filter
-from matplotlib.backends.backend_agg import FigureCanvasAgg as FigureCanvas
-from paddlehub.transforms.functional import *
-
-matplotlib.use('Agg')
+import paddlehub.vision.functional as F
 
 
 class Compose:
@@ -62,7 +50,7 @@ class Compose:
             im = cv2.cvtColor(im, cv2.COLOR_RGB2BGR)
 
         if self.is_permute:
-            im = permute(im)
+            im = F.permute(im)
 
         return im
 
@@ -73,7 +61,7 @@ class RandomHorizontalFlip:
 
     def __call__(self, im):
         if random.random() < self.prob:
-            im = horizontal_flip(im)
+            im = F.horizontal_flip(im)
         return im
 
 
@@ -83,13 +71,13 @@ class RandomVerticalFlip:
 
     def __call__(self, im):
         if random.random() < self.prob:
-            im = vertical_flip(im)
+            im = F.vertical_flip(im)
         return im
 
 
 class Resize:
     # The interpolation mode
-    interp_dict = {
+    interpolation_dict = {
         'NEAREST': cv2.INTER_NEAREST,
         'LINEAR': cv2.INTER_LINEAR,
         'CUBIC': cv2.INTER_CUBIC,
@@ -97,10 +85,10 @@ class Resize:
         'LANCZOS4': cv2.INTER_LANCZOS4
     }
 
-    def __init__(self, target_size=512, interp='LINEAR'):
-        self.interp = interp
-        if not (interp == "RANDOM" or interp in self.interp_dict):
-            raise ValueError("interp should be one of {}".format(self.interp_dict.keys()))
+    def __init__(self, target_size=512, interpolation='LINEAR'):
+        self.interpolation = interpolation
+        if not (interpolation == "RANDOM" or interpolation in self.interpolation_dict):
+            raise ValueError("interpolation should be one of {}".format(self.interpolation_dict.keys()))
         if isinstance(target_size, list) or isinstance(target_size, tuple):
             if len(target_size) != 2:
                 raise TypeError(
@@ -112,11 +100,11 @@ class Resize:
         self.target_size = target_size
 
     def __call__(self, im):
-        if self.interp == "RANDOM":
-            interp = random.choice(list(self.interp_dict.keys()))
+        if self.interpolation == "RANDOM":
+            interpolation = random.choice(list(self.interpolation_dict.keys()))
         else:
-            interp = self.interp
-        im = resize(im, self.target_size, self.interp_dict[interp])
+            interpolation = self.interpolation
+        im = F.resize(im, self.target_size, self.interpolation_dict[interpolation])
         return im
 
 
@@ -125,7 +113,7 @@ class ResizeByLong:
         self.long_size = long_size
 
     def __call__(self, im):
-        im = resize_long(im, self.long_size)
+        im = F.resize_long(im, self.long_size)
         return im
 
 
@@ -142,7 +130,7 @@ class ResizeRangeScaling:
             random_size = self.max_value
         else:
             random_size = int(np.random.uniform(self.min_value, self.max_value) + 0.5)
-        im = resize_long(im, random_size, cv2.INTER_LINEAR)
+        im = F.resize_long(im, random_size, cv2.INTER_LINEAR)
         return im
 
 
@@ -170,7 +158,7 @@ class ResizeStepScaling:
         w = int(round(scale_factor * im.shape[1]))
         h = int(round(scale_factor * im.shape[0]))
 
-        im = resize(im, (w, h), cv2.INTER_LINEAR)
+        im = F.resize(im, (w, h), cv2.INTER_LINEAR)
         return im
 
 
@@ -187,7 +175,7 @@ class Normalize:
     def __call__(self, im):
         mean = np.array(self.mean)[np.newaxis, np.newaxis, :]
         std = np.array(self.std)[np.newaxis, np.newaxis, :]
-        im = normalize(im, mean, std)
+        im = F.normalize(im, mean, std)
         return im
 
 
@@ -405,7 +393,7 @@ class RandomDistort:
             'hue': self.hue_prob
         }
         im = im.astype('uint8')
-        im = Image.fromarray(im)
+        im = PIL.Image.fromarray(im)
         for id in range(4):
             params = params_dict[ops[id].__name__]
             prob = prob_dict[ops[id].__name__]
@@ -592,7 +580,7 @@ class CenterCrop:
         self.crop_size = crop_size
 
     def __call__(self, img: np.ndarray):
-        img_width, img_height, chanel = img.shape
+        img_width, img_height, _ = img.shape
         crop_top = int((img_height - self.crop_size) / 2.)
         crop_left = int((img_width - self.crop_size) / 2.)
         return img[crop_left:crop_left + self.crop_size, crop_top:crop_top + self.crop_size, :]
@@ -622,14 +610,14 @@ class ResizeScaling:
 
     Args:
         target(int): Target image size.
-        interp(Callable): Interpolation method.
+        interpolation(Callable): Interpolation method.
     """
 
-    def __init__(self, target: int = 368, interp: Callable = cv2.INTER_CUBIC):
+    def __init__(self, target: int = 368, interpolation: Callable = cv2.INTER_CUBIC):
         self.target = target
-        self.interp = interp
+        self.interpolation = interpolation
 
     def __call__(self, img, scale_search):
         scale = scale_search * self.target / img.shape[0]
-        resize_img = cv2.resize(img, (0, 0), fx=scale, fy=scale, interpolation=self.interp)
+        resize_img = cv2.resize(img, (0, 0), fx=scale, fy=scale, interpolation=self.interpolation)
         return resize_img

paddlehub/transforms/functional.py → paddlehub/vision/utils.py

@@ -11,98 +11,13 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
 import os
 
-import cv2
 import paddle
-import matplotlib
+import PIL
 import numpy as np
-from pycocotools.coco import COCO
-from PIL import Image, ImageEnhance
-from matplotlib import pyplot as plt
-
-matplotlib.use('Agg')
-
-
-def normalize(im, mean, std):
-    im = im.astype(np.float32, copy=False) / 255.0
-    im -= mean
-    im /= std
-    return im
-
-
-def permute(im):
-    im = np.transpose(im, (2, 0, 1))
-    return im
-
-
-def resize(im, target_size=608, interp=cv2.INTER_LINEAR):
-    if isinstance(target_size, list) or isinstance(target_size, tuple):
-        w = target_size[0]
-        h = target_size[1]
-    else:
-        w = target_size
-        h = target_size
-    im = cv2.resize(im, (w, h), interpolation=interp)
-    return im
-
-
-def resize_long(im, long_size=224, interpolation=cv2.INTER_LINEAR):
-    value = max(im.shape[0], im.shape[1])
-    scale = float(long_size) / float(value)
-    resized_width = int(round(im.shape[1] * scale))
-    resized_height = int(round(im.shape[0] * scale))
-
-    im = cv2.resize(im, (resized_width, resized_height), interpolation=interpolation)
-    return im
-
-
-def horizontal_flip(im):
-    if len(im.shape) == 3:
-        im = im[:, ::-1, :]
-    elif len(im.shape) == 2:
-        im = im[:, ::-1]
-    return im
-
-
-def vertical_flip(im):
-    if len(im.shape) == 3:
-        im = im[::-1, :, :]
-    elif len(im.shape) == 2:
-        im = im[::-1, :]
-    return im
-
-
-def brightness(im, brightness_lower, brightness_upper):
-    brightness_delta = np.random.uniform(brightness_lower, brightness_upper)
-    im = ImageEnhance.Brightness(im).enhance(brightness_delta)
-    return im
-
-
-def contrast(im, contrast_lower, contrast_upper):
-    contrast_delta = np.random.uniform(contrast_lower, contrast_upper)
-    im = ImageEnhance.Contrast(im).enhance(contrast_delta)
-    return im
-
-
-def saturation(im, saturation_lower, saturation_upper):
-    saturation_delta = np.random.uniform(saturation_lower, saturation_upper)
-    im = ImageEnhance.Color(im).enhance(saturation_delta)
-    return im
-
-
-def hue(im, hue_lower, hue_upper):
-    hue_delta = np.random.uniform(hue_lower, hue_upper)
-    im = np.array(im.convert('HSV'))
-    im[:, :, 0] = im[:, :, 0] + hue_delta
-    im = Image.fromarray(im, mode='HSV').convert('RGB')
-    return im
-
-
-def rotate(im, rotate_lower, rotate_upper):
-    rotate_delta = np.random.uniform(rotate_lower, rotate_upper)
-    im = im.rotate(int(rotate_delta))
-    return im
+import matplotlib as plt
 
 
 def is_image_file(filename: str) -> bool:
@@ -114,13 +29,13 @@ def is_image_file(filename: str) -> bool:
 def get_img_file(dir_name: str) -> list:
     '''Get all image file paths in several directories which have the same parent directory.'''
     images = []
-    for parent, dirnames, filenames in os.walk(dir_name):
+    for parent, _, filenames in os.walk(dir_name):
         for filename in filenames:
             if not is_image_file(filename):
                 continue
             img_path = os.path.join(parent, filename)
             images.append(img_path)
-    images.sort()
+
     return images
 
 
@@ -188,7 +103,7 @@ def draw_boxes_on_image(image_path: str,
                         score_thresh: float = 0.5,
                         save_path: str = 'result'):
     """Draw boxes on images."""
-    image = np.array(Image.open(image_path))
+    image = np.array(PIL.Image.open(image_path))
     plt.figure()
     _, ax = plt.subplots(1)
     ax.imshow(image)
@@ -230,16 +145,9 @@ def draw_boxes_on_image(image_path: str,
     plt.close('all')
 
 
-def img_shape(img_path: str):
-    """Get image shape."""
-    im = cv2.imread(img_path)
-    im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
-    h, w, c = im.shape
-    return h, w, c
-
-
 def get_label_infos(file_list: str):
     """Get label names by corresponding category ids."""
+    from pycocotools.coco import COCO
     map_label = COCO(file_list)
     label_names = []
     categories = map_label.loadCats(map_label.getCatIds())