add imagenet preprocessing example and more functions for image_reader

python/examples/imagenet/image_rpc_client.py

@@ -13,22 +13,24 @@
 # limitations under the License.
 
 import sys
-from image_reader import ImageReader
 from paddle_serving_client import Client
+from paddle_serving_app.reader import Sequential, File2Image, Resize, CenterCrop, RGB2BGR, Transpose, Div, Normalize
 import time
 
 client = Client()
 client.load_client_config(sys.argv[1])
 client.connect(["127.0.0.1:9393"])
-reader = ImageReader()
+
+seq = Sequential([
+    File2Image(), Resize(256), CenterCrop(224), RGB2BGR(), Transpose((2, 0, 1)),
+    Div(255), Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
+])
+print(seq)
 
 start = time.time()
+image_file = "daisy.jpg"
 for i in range(1000):
-    with open("./data/n01440764_10026.JPEG", "rb") as f:
-        img = f.read()
-    img = reader.process_image(img)
+    img = seq(image_file)
     fetch_map = client.predict(feed={"image": img}, fetch=["score"])
 end = time.time()
 print(end - start)
-
-#print(fetch_map["score"])

python/paddle_serving_app/reader/__init__.py

@@ -11,4 +11,4 @@
 # 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.
-from .image_reader import ImageReader, File2Image, URL2Image, Sequential, Normalize, CenterCrop, Resize
+from .image_reader import ImageReader, File2Image, URL2Image, Sequential, Normalize, CenterCrop, Resize, Transpose, Div, RGB2BGR, BGR2RGB

python/paddle_serving_app/reader/functional.py

@@ -16,14 +16,17 @@ import cv2
 import numpy as np
 
 
+def transpose(img, transpose_target):
+    img = img.transpose(transpose_target)
+    return img
+
+
 def normalize(img, mean, std):
     # need to optimize here
-    img = img.astype('float32').transpose((2, 0, 1)) / 255
     img_mean = np.array(mean).reshape((3, 1, 1))
     img_std = np.array(std).reshape((3, 1, 1))
     img -= img_mean
     img /= img_std
-    img = img.transpose((1, 2, 0))
     return img
 
 
@@ -47,7 +50,7 @@ def resize(img, target_size, interpolation):
         resized_width = target_size[0]
         resized_height = target_size[1]
     else:
-        percent = float(target_size) / min(img.shape[1], img.shape[2])
+        percent = float(target_size) / min(img.shape[0], img.shape[1])
         resized_width = int(round(img.shape[1] * percent))
         resized_height = int(round(img.shape[0] * percent))
     if interpolation:
@@ -55,5 +58,4 @@ def resize(img, target_size, interpolation):
             img, (resized_width, resized_height), interpolation=interpolation)
     else:
         resized = cv2.resize(img, (resized_width, resized_height))
-    print(resized.shape)
     return resized

python/paddle_serving_app/reader/image_reader.py

@@ -17,7 +17,7 @@ import numpy as np
 import base64
 import functional as F
 
-_cv2_interpolation_to_str = {cv2.INTER_LINEAR: "cv2.INTER_LINEAR"}
+_cv2_interpolation_to_str = {cv2.INTER_LINEAR: "cv2.INTER_LINEAR", None: "None"}
 
 
 class Sequential(object):
@@ -51,6 +51,28 @@ class Sequential(object):
         return format_string_
 
 
+class RGB2BGR(object):
+    def __init__(self):
+        pass
+
+    def __call__(self, img):
+        return img[:, :, ::-1]
+
+    def __repr__(self):
+        return self.__class__.__name__ + "()"
+
+
+class BGR2RGB(object):
+    def __init__(self):
+        pass
+
+    def __call__(self, img):
+        return img[:, :, ::-1]
+
+    def __repr__(self):
+        return self.__class__.__name__ + "()"
+
+
 class File2Image(object):
     def __init__(self):
         pass
@@ -81,6 +103,40 @@ class URL2Image(object):
         return self.__class__.__name__ + "()"
 
 
+class Base64ToImage(object):
+    def __init__(self):
+        pass
+
+    def __call__(self, img_base64):
+        img = base64.b64decode(img_base64)
+        return img
+
+    def __repr__(self):
+        return self.__class__.__name__ + "()"
+
+
+class Div(object):
+    """ divide by some float number """
+
+    def __init__(self, value):
+        self.value = value
+
+    def __call__(self, img):
+        """
+        Args:
+            img (numpy array): (int8 numpy array)
+
+        Returns:
+            img (numpy array): (float32 numpy array)
+        """
+        img = img.astype('float32') / self.value
+
+        return img
+
+    def __repr__(self):
+        return self.__class__.__name__ + "({})".format(self.value)
+
+
 class Normalize(object):
     """Normalize a tensor image with mean and standard deviation.
     Given mean: ``(M1,...,Mn)`` and std: ``(S1,..,Sn)`` for ``n`` channels, this transform
@@ -115,6 +171,27 @@ class Normalize(object):
                                                                       self.std)
 
 
+class Lambda(object):
+    """Apply a user-defined lambda as a transform.
+       Very shame to just copy from 
+       https://github.com/pytorch/vision/blob/master/torchvision/transforms/transforms.py#L301
+
+    Args:
+        lambd (function): Lambda/function to be used for transform.
+    """
+
+    def __init__(self, lambd):
+        assert callable(lambd), repr(type(lambd)
+                                     .__name__) + " object is not callable"
+        self.lambd = lambd
+
+    def __call__(self, img):
+        return self.lambd(img)
+
+    def __repr__(self):
+        return self.__class__.__name__ + '()'
+
+
 class CenterCrop(object):
     """Crops the given Image at the center.
 
@@ -154,18 +231,32 @@ class Resize(object):
             ``PIL.Image.BILINEAR``
     """
 
-    def __init__(self, size, interpolation=cv2.INTER_LINEAR):
+    def __init__(self, size, interpolation=None):
         self.size = size
         self.interpolation = interpolation
 
     def __call__(self, img):
         return F.resize(img, self.size, self.interpolation)
 
-    def __repr__(self, img):
+    def __repr__(self):
         return self.__class__.__name__ + '(size={0}, interpolation={1})'.format(
             self.size, _cv2_interpolation_to_str[self.interpolation])
 
 
+class Transpose(object):
+    def __init__(self, transpose_target):
+        self.transpose_target = transpose_target
+
+    def __call__(self, img):
+        return F.transpose(img, self.transpose_target)
+        return img
+
+    def __repr__(self):
+        format_string = self.__class__.__name__ + \
+                        "({})".format(self.transpose_target)
+        return format_string
+
+
 class ImageReader():
     def __init__(self,
                  image_shape=[3, 224, 224],