From b806e3fdbbc390bd7a095d65ebcfb5399d7afb81 Mon Sep 17 00:00:00 2001
From: felixhjh <852142024@qq.com>
Date: Mon, 18 Oct 2021 07:19:12 +0000
Subject: [PATCH 1/7] modify detection examples preprocess

---
 python/examples/cascade_rcnn/test_client.py   |  17 +-
 .../faster_rcnn_hrnetv2p_w18_1x/README.md     |   2 +-
 .../test_client.py                            |  15 +-
 .../test_client.py                            |  19 +-
 .../fcos_dcn_r50_fpn_1x_coco/test_client.py   |  21 ++-
 .../ppyolo_r50vd_dcn_1x_coco/test_client.py   |  16 +-
 .../ssd_vgg16_300_240e_voc/test_client.py     |  22 ++-
 .../yolov3_darknet53_270e_coco/test_client.py |  17 +-
 python/examples/yolov4/test_client.py         |   2 +-
 .../paddle_serving_app/reader/image_reader.py | 177 +++++++++++++-----
 10 files changed, 203 insertions(+), 105 deletions(-)

diff --git a/python/examples/cascade_rcnn/test_client.py b/python/examples/cascade_rcnn/test_client.py
index 5dbdd5567..cd90e5f15 100644
--- a/python/examples/cascade_rcnn/test_client.py
+++ b/python/examples/cascade_rcnn/test_client.py
@@ -17,27 +17,30 @@
 from paddle_serving_client import Client
 from paddle_serving_app.reader import *
 import cv2
+
 preprocess = Sequential([
-    File2Image(), BGR2RGB(), Resize(
-        (608, 608), interpolation=cv2.INTER_LINEAR), Div(255.0), Transpose(
-            (2, 0, 1))
+        File2Image(),
+        Resize((800, 1333), True, interpolation=2), 
+        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        Transpose((2,0,1)),
+        PadStride(32)
 ])
 
-postprocess = RCNNPostprocess("label_list.txt", "output", [608, 608])
+postprocess = RCNNPostprocess("label_list.txt", "output")
 client = Client()
 
 client.load_client_config("serving_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9292'])
 
-im = preprocess('000000570688.jpg')
+im, im_info = preprocess(sys.argv[1])
 fetch_map = client.predict(
     feed={
         "image": im,
         "im_shape": np.array(list(im.shape[1:])).reshape(-1),
-        "scale_factor": np.array([1.0, 1.0]).reshape(-1),
+        "scale_factor": im_info['scale_factor'],
     },
     fetch=["save_infer_model/scale_0.tmp_1"],
     batch=False)
 print(fetch_map)
-fetch_map["image"] = '000000570688.jpg'
+fetch_map["image"] = sys.argv[1]
 postprocess(fetch_map)
diff --git a/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/README.md b/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/README.md
index efd8b9f0f..3c0fb8dbe 100644
--- a/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/README.md
+++ b/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/README.md
@@ -9,7 +9,7 @@ wget --no-check-certificate https://paddle-serving.bj.bcebos.com/pddet_demo/fast
 
 ### Start the service
 ```
-tar xf faster_rcnn_hrnetv2p_w18_1x.tar
+tar xf faster_rcnn_hrnetv2p_w18_1x.tar.gz
 python3 -m paddle_serving_server.serve --model serving_server --port 9494 --gpu_ids 0
 ```
 
diff --git a/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/test_client.py b/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/test_client.py
index 22cc1451c..b68a0c7b4 100644
--- a/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/test_client.py
+++ b/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/test_client.py
@@ -17,24 +17,27 @@
 from paddle_serving_client import Client
 from paddle_serving_app.reader import *
 import cv2
+
 preprocess = Sequential([
-    File2Image(), BGR2RGB(), Resize(
-        (608, 608), interpolation=cv2.INTER_LINEAR), Div(255.0), Transpose(
-            (2, 0, 1))
+        File2Image(),
+        Resize((800, 1333), True, interpolation=2), 
+        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        Transpose((2,0,1)),
+        PadStride(32)
 ])
 
-postprocess = RCNNPostprocess("label_list.txt", "output", [608, 608])
+postprocess = RCNNPostprocess("label_list.txt", "output")
 client = Client()
 
 client.load_client_config("serving_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9494'])
 
-im = preprocess(sys.argv[1])
+im, im_info = preprocess(sys.argv[1])
 fetch_map = client.predict(
     feed={
         "image": im,
         "im_shape": np.array(list(im.shape[1:])).reshape(-1),
-        "scale_factor": np.array([1.0, 1.0]).reshape(-1),
+        "scale_factor": im_info['scale_factor'],
     },
     fetch=["save_infer_model/scale_0.tmp_1"],
     batch=False)
diff --git a/python/examples/detection/faster_rcnn_r50_fpn_1x_coco/test_client.py b/python/examples/detection/faster_rcnn_r50_fpn_1x_coco/test_client.py
index a81f53dce..5d5e0f2b1 100644
--- a/python/examples/detection/faster_rcnn_r50_fpn_1x_coco/test_client.py
+++ b/python/examples/detection/faster_rcnn_r50_fpn_1x_coco/test_client.py
@@ -12,15 +12,19 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from paddle_serving_client import Client
-from paddle_serving_app.reader import *
 import sys
 import numpy as np
+from paddle_serving_client import Client
+from paddle_serving_app.reader import *
+import cv2
 
 preprocess = Sequential([
-    File2Image(), BGR2RGB(), Div(255.0),
-    Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], False),
-    Resize(640, 640), Transpose((2, 0, 1))
+        File2Image(),
+        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        Resize(
+        (800, 1333), True, interpolation=cv2.INTER_LINEAR), 
+        Transpose((2,0,1)),
+        PadStride(128)
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
@@ -29,15 +33,14 @@
 client.load_client_config("serving_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9494'])
 
-im = preprocess(sys.argv[1])
+im, im_info = preprocess(sys.argv[1])
 fetch_map = client.predict(
     feed={
         "image": im,
         "im_shape": np.array(list(im.shape[1:])).reshape(-1),
-        "scale_factor": np.array([1.0, 1.0]).reshape(-1),
+        "scale_factor": im_info['scale_factor'],
     },
     fetch=["save_infer_model/scale_0.tmp_1"],
     batch=False)
-print(fetch_map)
 fetch_map["image"] = sys.argv[1]
 postprocess(fetch_map)
diff --git a/python/examples/detection/fcos_dcn_r50_fpn_1x_coco/test_client.py b/python/examples/detection/fcos_dcn_r50_fpn_1x_coco/test_client.py
index bf5504105..f7cdc745b 100644
--- a/python/examples/detection/fcos_dcn_r50_fpn_1x_coco/test_client.py
+++ b/python/examples/detection/fcos_dcn_r50_fpn_1x_coco/test_client.py
@@ -12,15 +12,19 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from paddle_serving_client import Client
-from paddle_serving_app.reader import *
 import sys
 import numpy as np
+from paddle_serving_client import Client
+from paddle_serving_app.reader import *
+import cv2
 
 preprocess = Sequential([
-    File2Image(), BGR2RGB(), Div(255.0),
-    Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], False),
-    Resize(640, 640), Transpose((2, 0, 1))
+        File2Image(),
+        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        Resize(
+        (800, 1333), True, interpolation=cv2.INTER_LINEAR), 
+        Transpose((2,0,1)),
+        PadStride(128)
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
@@ -29,12 +33,13 @@
 client.load_client_config("serving_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9494'])
 
-im = preprocess(sys.argv[1])
+im, im_info = preprocess(sys.argv[1])
 fetch_map = client.predict(
     feed={
         "image": im,
-        "scale_factor": np.array([1.0, 1.0]).reshape(-1),
+        "scale_factor": im_info['scale_factor'],
     },
     fetch=["save_infer_model/scale_0.tmp_1"],
     batch=False)
-print(fetch_map)
+fetch_map["image"] = sys.argv[1]
+postprocess(fetch_map)
diff --git a/python/examples/detection/ppyolo_r50vd_dcn_1x_coco/test_client.py b/python/examples/detection/ppyolo_r50vd_dcn_1x_coco/test_client.py
index 22cc1451c..f26e13446 100644
--- a/python/examples/detection/ppyolo_r50vd_dcn_1x_coco/test_client.py
+++ b/python/examples/detection/ppyolo_r50vd_dcn_1x_coco/test_client.py
@@ -17,27 +17,29 @@
 from paddle_serving_client import Client
 from paddle_serving_app.reader import *
 import cv2
+
 preprocess = Sequential([
-    File2Image(), BGR2RGB(), Resize(
-        (608, 608), interpolation=cv2.INTER_LINEAR), Div(255.0), Transpose(
-            (2, 0, 1))
+        File2Image(),
+        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        Resize(
+        (608, 608), False, interpolation=2), 
+        Transpose((2,0,1))
 ])
 
-postprocess = RCNNPostprocess("label_list.txt", "output", [608, 608])
+postprocess = RCNNPostprocess("label_list.txt", "output")
 client = Client()
 
 client.load_client_config("serving_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9494'])
 
-im = preprocess(sys.argv[1])
+im, im_info = preprocess(sys.argv[1])
 fetch_map = client.predict(
     feed={
         "image": im,
         "im_shape": np.array(list(im.shape[1:])).reshape(-1),
-        "scale_factor": np.array([1.0, 1.0]).reshape(-1),
+        "scale_factor": im_info['scale_factor'],
     },
     fetch=["save_infer_model/scale_0.tmp_1"],
     batch=False)
-print(fetch_map)
 fetch_map["image"] = sys.argv[1]
 postprocess(fetch_map)
diff --git a/python/examples/detection/ssd_vgg16_300_240e_voc/test_client.py b/python/examples/detection/ssd_vgg16_300_240e_voc/test_client.py
index 3409ef885..f7cdc745b 100644
--- a/python/examples/detection/ssd_vgg16_300_240e_voc/test_client.py
+++ b/python/examples/detection/ssd_vgg16_300_240e_voc/test_client.py
@@ -12,15 +12,19 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from paddle_serving_client import Client
-from paddle_serving_app.reader import *
 import sys
 import numpy as np
+from paddle_serving_client import Client
+from paddle_serving_app.reader import *
+import cv2
 
 preprocess = Sequential([
-    File2Image(), BGR2RGB(),
-    Normalize([123.675, 116.28, 103.53], [58.395, 57.12, 57.375], False),
-    Resize((512, 512)), Transpose((2, 0, 1))
+        File2Image(),
+        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        Resize(
+        (800, 1333), True, interpolation=cv2.INTER_LINEAR), 
+        Transpose((2,0,1)),
+        PadStride(128)
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
@@ -29,13 +33,13 @@
 client.load_client_config("serving_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9494'])
 
-im = preprocess(sys.argv[1])
+im, im_info = preprocess(sys.argv[1])
 fetch_map = client.predict(
     feed={
         "image": im,
-        "im_shape": np.array([512, 512]),
-        "scale_factor": np.array([1.0, 1.0]).reshape(-1),
+        "scale_factor": im_info['scale_factor'],
     },
     fetch=["save_infer_model/scale_0.tmp_1"],
     batch=False)
-print(fetch_map)
+fetch_map["image"] = sys.argv[1]
+postprocess(fetch_map)
diff --git a/python/examples/detection/yolov3_darknet53_270e_coco/test_client.py b/python/examples/detection/yolov3_darknet53_270e_coco/test_client.py
index 22cc1451c..5d5e0f2b1 100644
--- a/python/examples/detection/yolov3_darknet53_270e_coco/test_client.py
+++ b/python/examples/detection/yolov3_darknet53_270e_coco/test_client.py
@@ -17,27 +17,30 @@
 from paddle_serving_client import Client
 from paddle_serving_app.reader import *
 import cv2
+
 preprocess = Sequential([
-    File2Image(), BGR2RGB(), Resize(
-        (608, 608), interpolation=cv2.INTER_LINEAR), Div(255.0), Transpose(
-            (2, 0, 1))
+        File2Image(),
+        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        Resize(
+        (800, 1333), True, interpolation=cv2.INTER_LINEAR), 
+        Transpose((2,0,1)),
+        PadStride(128)
 ])
 
-postprocess = RCNNPostprocess("label_list.txt", "output", [608, 608])
+postprocess = RCNNPostprocess("label_list.txt", "output")
 client = Client()
 
 client.load_client_config("serving_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9494'])
 
-im = preprocess(sys.argv[1])
+im, im_info = preprocess(sys.argv[1])
 fetch_map = client.predict(
     feed={
         "image": im,
         "im_shape": np.array(list(im.shape[1:])).reshape(-1),
-        "scale_factor": np.array([1.0, 1.0]).reshape(-1),
+        "scale_factor": im_info['scale_factor'],
     },
     fetch=["save_infer_model/scale_0.tmp_1"],
     batch=False)
-print(fetch_map)
 fetch_map["image"] = sys.argv[1]
 postprocess(fetch_map)
diff --git a/python/examples/yolov4/test_client.py b/python/examples/yolov4/test_client.py
index dfcd58610..3db9b3ec3 100644
--- a/python/examples/yolov4/test_client.py
+++ b/python/examples/yolov4/test_client.py
@@ -29,7 +29,7 @@
 client.load_client_config("yolov4_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9393'])
 
-im = preprocess(sys.argv[1])
+im, im_info = preprocess(sys.argv[1])
 fetch_map = client.predict(
     feed={
         "image": im,
diff --git a/python/paddle_serving_app/reader/image_reader.py b/python/paddle_serving_app/reader/image_reader.py
index ca97af94f..3a323a4b3 100644
--- a/python/paddle_serving_app/reader/image_reader.py
+++ b/python/paddle_serving_app/reader/image_reader.py
@@ -484,10 +484,15 @@ class Sequential(object):
     def __init__(self, transforms):
         self.transforms = transforms
 
-    def __call__(self, img):
+    def __call__(self, im):
+        im_info = {
+        'scale_factor': np.array(
+            [1., 1.], dtype=np.float32),
+        'im_shape': None,
+        }
         for t in self.transforms:
-            img = t(img)
-        return img
+            im, im_info = t(im, im_info)
+        return im, im_info
 
     def __repr__(self):
         format_string_ = self.__class__.__name__ + '('
@@ -513,8 +518,8 @@ class BGR2RGB(object):
     def __init__(self):
         pass
 
-    def __call__(self, img):
-        return img[:, :, ::-1]
+    def __call__(self, img, img_info = False):
+        return img[:, :, ::-1], img_info
 
     def __repr__(self):
         return self.__class__.__name__ + "()"
@@ -537,7 +542,7 @@ class File2Image(object):
     def __init__(self):
         pass
 
-    def __call__(self, img_path):
+    def __call__(self, img_path, im_info):
         if py_version == 2:
             fin = open(img_path)
         else:
@@ -545,13 +550,16 @@ def __call__(self, img_path):
         sample = fin.read()
         data = np.fromstring(sample, np.uint8)
         img = cv2.imdecode(data, cv2.IMREAD_COLOR)
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
         '''
         img = cv2.imread(img_path, -1)
         channels = img.shape[2]
         ori_h = img.shape[0]
         ori_w = img.shape[1]
         '''
-        return img
+        im_info['im_shape'] = np.array(img.shape[:2], dtype=np.float32)
+        im_info['scale_factor'] = np.array([1., 1.], dtype=np.float32)
+        return img, im_info
 
     def __repr__(self):
         return self.__class__.__name__ + "()"
@@ -592,7 +600,7 @@ class Div(object):
     def __init__(self, value):
         self.value = value
 
-    def __call__(self, img):
+    def __call__(self, img, img_info = False):
         """
         Args:
             img (numpy array): (int8 numpy array)
@@ -602,7 +610,7 @@ def __call__(self, img):
         """
         img = img.astype('float32') / self.value
 
-        return img
+        return img, img_info
 
     def __repr__(self):
         return self.__class__.__name__ + "({})".format(self.value)
@@ -620,23 +628,33 @@ class Normalize(object):
     Args:
         mean (sequence): Sequence of means for each channel.
         std (sequence): Sequence of standard deviations for each channel.
+        is_scale (bool): whether need im / 255
 
     """
 
-    def __init__(self, mean, std, channel_first=False):
+    def __init__(self, mean, std, is_scale=True):
         self.mean = mean
         self.std = std
-        self.channel_first = channel_first
+        self.is_scale = is_scale
 
-    def __call__(self, img):
+    def __call__(self, im, im_info = False):
         """
         Args:
-            img (numpy array): (C, H, W) to be normalized.
-
+            im (np.ndarray): image (np.ndarray)
+            im_info (dict): info of image
         Returns:
-            Tensor: Normalized Tensor image.
+            im (np.ndarray):  processed image (np.ndarray)
+            im_info (dict): info of processed image
         """
-        return F.normalize(img, self.mean, self.std, self.channel_first)
+        im = im.astype(np.float32, copy=False)
+        mean = np.array(self.mean)[np.newaxis, np.newaxis, :]
+        std = np.array(self.std)[np.newaxis, np.newaxis, :]
+
+        if self.is_scale:
+            im = im / 255.0
+        im -= mean
+        im /= std
+        return im, im_info
 
     def __repr__(self):
         return self.__class__.__name__ + '(mean={0}, std={1})'.format(self.mean,
@@ -690,26 +708,72 @@ def __repr__(self):
         return self.__class__.__name__ + '(size={0})'.format(self.size)
 
 
-class Resize(object):
-    """Resize the input numpy array Image to the given size.
 
+class Resize(object):
+    """resize image by target_size and max_size
     Args:
-        size (sequence or int): Desired output size. If size is a sequence like
-            (w, h), output size will be matched to this. If size is an int,
-            smaller edge of the image will be matched to this number.
-            i.e, if height > width, then image will be rescaled to
-            (size * height / width, size)
-        interpolation (int, optional): Desired interpolation. Default is
-            ``None``
+        target_size (int): the target size of image
+        keep_ratio (bool): whether keep_ratio or not, default true
+        interp (int): method of resize
     """
 
-    def __init__(self, size, max_size=2147483647, interpolation=None):
-        self.size = size
-        self.max_size = max_size
+    def __init__(self, target_size, keep_ratio=True, interpolation=cv2.INTER_LINEAR):
+        if isinstance(target_size, int):
+            target_size = [target_size, target_size]
+        self.target_size = target_size
+        self.keep_ratio = keep_ratio
         self.interpolation = interpolation
 
-    def __call__(self, img):
-        return F.resize(img, self.size, self.max_size, self.interpolation)
+    def __call__(self, im, im_info):
+        """
+        Args:
+            im (np.ndarray): image (np.ndarray)
+            im_info (dict): info of image
+        Returns:
+            im (np.ndarray):  processed image (np.ndarray)
+            im_info (dict): info of processed image
+        """
+        assert len(self.target_size) == 2
+        assert self.target_size[0] > 0 and self.target_size[1] > 0
+        im_channel = im.shape[2]
+        im_scale_y, im_scale_x = self.generate_scale(im)
+        im = cv2.resize(
+            im,
+            None,
+            None,
+            fx=im_scale_x,
+            fy=im_scale_y,
+            interpolation=self.interpolation)
+        im_info['im_shape'] = np.array(im.shape[:2]).astype('float32')
+        im_info['scale_factor'] = np.array(
+            [im_scale_y, im_scale_x]).astype('float32')
+        return im, im_info
+
+    def generate_scale(self, im):
+        """
+        Args:
+            im (np.ndarray): image (np.ndarray)
+        Returns:
+            im_scale_x: the resize ratio of X
+            im_scale_y: the resize ratio of Y
+        """
+        origin_shape = im.shape[:2]
+        im_c = im.shape[2]
+        if self.keep_ratio:
+            im_size_min = np.min(origin_shape)
+            im_size_max = np.max(origin_shape)
+            target_size_min = np.min(self.target_size)
+            target_size_max = np.max(self.target_size)
+            im_scale = float(target_size_min) / float(im_size_min)
+            if np.round(im_scale * im_size_max) > target_size_max:
+                im_scale = float(target_size_max) / float(im_size_max)
+            im_scale_x = im_scale
+            im_scale_y = im_scale
+        else:
+            resize_h, resize_w = self.target_size
+            im_scale_y = resize_h / float(origin_shape[0])
+            im_scale_x = resize_w / float(origin_shape[1])
+        return im_scale_y, im_scale_x
 
     def __repr__(self):
         return self.__class__.__name__ + '(size={0}, max_size={1}, interpolation={2})'.format(
@@ -717,6 +781,35 @@ def __repr__(self):
             _cv2_interpolation_to_str[self.interpolation])
 
 
+class PadStride(object):
+    """ padding image for model with FPN, instead PadBatch(pad_to_stride) in original config
+    Args:
+        stride (bool): model with FPN need image shape % stride == 0
+    """
+
+    def __init__(self, stride=0):
+        self.coarsest_stride = stride
+
+    def __call__(self, im, im_info = False):
+        """
+        Args:
+            im (np.ndarray): image (np.ndarray)
+            im_info (dict): info of image
+        Returns:
+            im (np.ndarray):  processed image (np.ndarray)
+            im_info (dict): info of processed image
+        """
+        coarsest_stride = self.coarsest_stride
+        if coarsest_stride <= 0:
+            return im
+        im_c, im_h, im_w = im.shape
+        pad_h = int(np.ceil(float(im_h) / coarsest_stride) * coarsest_stride)
+        pad_w = int(np.ceil(float(im_w) / coarsest_stride) * coarsest_stride)
+        padding_im = np.zeros((im_c, pad_h, pad_w), dtype=np.float32)
+        padding_im[:, :im_h, :im_w] = im
+        return padding_im, im_info
+
+
 class ResizeByFactor(object):
     """Resize the input numpy array Image to a size multiple of factor which is usually required by a network
 
@@ -768,31 +861,13 @@ def __repr__(self):
             self.factor, self.max_side_len)
 
 
-class PadStride(object):
-    def __init__(self, stride):
-        self.coarsest_stride = stride
-
-    def __call__(self, img):
-        coarsest_stride = self.coarsest_stride
-        if coarsest_stride == 0:
-            return img
-        im_c, im_h, im_w = img.shape
-        pad_h = int(np.ceil(float(im_h) / coarsest_stride) * coarsest_stride)
-        pad_w = int(np.ceil(float(im_w) / coarsest_stride) * coarsest_stride)
-        padding_im = np.zeros((im_c, pad_h, pad_w), dtype=np.float32)
-        padding_im[:, :im_h, :im_w] = img
-        im_info = {}
-        im_info['resize_shape'] = padding_im.shape[1:]
-        return padding_im
-
-
 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 __call__(self, im, im_info = False):
+        im = F.transpose(im, self.transpose_target)
+        return im, im_info
 
     def __repr__(self):
         format_string = self.__class__.__name__ + \

From 1659fad659885d43d005c7fb3f65b24a3f7d6331 Mon Sep 17 00:00:00 2001
From: huangjianhui <852142024@qq.com>
Date: Mon, 18 Oct 2021 16:08:34 +0800
Subject: [PATCH 2/7] Update image_reader.py

---
 python/paddle_serving_app/reader/image_reader.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/python/paddle_serving_app/reader/image_reader.py b/python/paddle_serving_app/reader/image_reader.py
index 3a323a4b3..380079c28 100644
--- a/python/paddle_serving_app/reader/image_reader.py
+++ b/python/paddle_serving_app/reader/image_reader.py
@@ -637,7 +637,7 @@ def __init__(self, mean, std, is_scale=True):
         self.std = std
         self.is_scale = is_scale
 
-    def __call__(self, im, im_info = False):
+    def __call__(self, im, im_info = None):
         """
         Args:
             im (np.ndarray): image (np.ndarray)
@@ -790,7 +790,7 @@ class PadStride(object):
     def __init__(self, stride=0):
         self.coarsest_stride = stride
 
-    def __call__(self, im, im_info = False):
+    def __call__(self, im, im_info = None):
         """
         Args:
             im (np.ndarray): image (np.ndarray)
@@ -865,7 +865,7 @@ class Transpose(object):
     def __init__(self, transpose_target):
         self.transpose_target = transpose_target
 
-    def __call__(self, im, im_info = False):
+    def __call__(self, im, im_info = None):
         im = F.transpose(im, self.transpose_target)
         return im, im_info
 

From da0f7b12e82be25aef40b3d6878b54c5ccd2e978 Mon Sep 17 00:00:00 2001
From: huangjianhui <852142024@qq.com>
Date: Mon, 18 Oct 2021 17:19:46 +0800
Subject: [PATCH 3/7] Update test_client.py

---
 .../ttfnet_darknet53_1x_coco/test_client.py   | 21 ++++++++++++-------
 1 file changed, 13 insertions(+), 8 deletions(-)

diff --git a/python/examples/detection/ttfnet_darknet53_1x_coco/test_client.py b/python/examples/detection/ttfnet_darknet53_1x_coco/test_client.py
index 59024d010..114f43c1b 100644
--- a/python/examples/detection/ttfnet_darknet53_1x_coco/test_client.py
+++ b/python/examples/detection/ttfnet_darknet53_1x_coco/test_client.py
@@ -11,16 +11,18 @@
 # 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 paddle_serving_client import Client
-from paddle_serving_app.reader import *
 import sys
 import numpy as np
+from paddle_serving_client import Client
+from paddle_serving_app.reader import *
+import cv2
 
 preprocess = Sequential([
-    File2Image(), BGR2RGB(), 
-    Normalize([123.675, 116.28, 103.53], [58.395, 57.12, 57.375], False),
-    Resize((512, 512)), Transpose((2, 0, 1))
+        File2Image(),
+        Resize(
+        (512, 512), False, interpolation=cv2.INTER_LINEAR),
+        Normalize([123.675, 116.28, 103.53], [58.395, 57.12, 57.375], False),
+        Transpose((2,0,1))
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
@@ -29,11 +31,14 @@
 client.load_client_config("serving_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9494'])
 
-im = preprocess(sys.argv[1])
+im, im_info = preprocess(sys.argv[1])
+
+
 fetch_map = client.predict(
     feed={
         "image": im,
-        "scale_factor": np.array([1.0, 1.0]).reshape(-1),
+        "im_shape": np.array(list(im.shape[1:])).reshape(-1),
+        "scale_factor": im_info['scale_factor'],
     },
     fetch=["save_infer_model/scale_0.tmp_1"],
     batch=False)

From 26ff9e79e9af9b2276624685bf007f44915bbc86 Mon Sep 17 00:00:00 2001
From: huangjianhui <852142024@qq.com>
Date: Mon, 18 Oct 2021 18:19:06 +0800
Subject: [PATCH 4/7] Update README.md

---
 python/examples/detection/ttfnet_darknet53_1x_coco/README.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/examples/detection/ttfnet_darknet53_1x_coco/README.md b/python/examples/detection/ttfnet_darknet53_1x_coco/README.md
index d6ffb912c..adf5de2ab 100644
--- a/python/examples/detection/ttfnet_darknet53_1x_coco/README.md
+++ b/python/examples/detection/ttfnet_darknet53_1x_coco/README.md
@@ -4,7 +4,7 @@
 
 ### Get Model
 ```
-wget --no-check-certificate https://paddle-serving.bj.bcebos.com/pddet_demo/2.0/ttfnet_darknet53_1x_coco.tar
+wget --no-check-certificate https://paddle-serving.bj.bcebos.com/pddet_demo/ttfnet_darknet53_1x_coco.tar
 ```
 
 ### Start the service

From 601e2beb7ae6a242f682420837271247d1ee76ce Mon Sep 17 00:00:00 2001
From: huangjianhui <852142024@qq.com>
Date: Mon, 18 Oct 2021 18:19:39 +0800
Subject: [PATCH 5/7] Update README_CN.md

---
 python/examples/detection/ttfnet_darknet53_1x_coco/README_CN.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/examples/detection/ttfnet_darknet53_1x_coco/README_CN.md b/python/examples/detection/ttfnet_darknet53_1x_coco/README_CN.md
index 7a1d3d1ab..7a2c86096 100644
--- a/python/examples/detection/ttfnet_darknet53_1x_coco/README_CN.md
+++ b/python/examples/detection/ttfnet_darknet53_1x_coco/README_CN.md
@@ -4,7 +4,7 @@
 
 ## 获得模型
 ```
-wget --no-check-certificate https://paddle-serving.bj.bcebos.com/pddet_demo/2.0/ttfnet_darknet53_1x_coco.tar
+wget --no-check-certificate https://paddle-serving.bj.bcebos.com/pddet_demo/ttfnet_darknet53_1x_coco.tar
 ```
 
 

From cfdc087f66724bb9fd41827787f5be06cb38dde8 Mon Sep 17 00:00:00 2001
From: felixhjh <852142024@qq.com>
Date: Tue, 19 Oct 2021 03:20:00 +0000
Subject: [PATCH 6/7] modify detection examples preprocess fix detection
 examples

---
 python/examples/cascade_rcnn/README.md        |   2 +-
 python/examples/cascade_rcnn/README_CN.md     |   2 +-
 python/examples/cascade_rcnn/test_client.py   |  12 +-
 .../faster_rcnn_hrnetv2p_w18_1x/README_CN.md  |   2 +-
 .../test_client.py                            |  12 +-
 .../test_client.py                            |  12 +-
 .../fcos_dcn_r50_fpn_1x_coco/test_client.py   |  14 +-
 .../ppyolo_r50vd_dcn_1x_coco/test_client.py   |  10 +-
 .../ssd_vgg16_300_240e_voc/label_list.txt     |  88 ++------
 .../ssd_vgg16_300_240e_voc/test_client.py     |  16 +-
 .../ttfnet_darknet53_1x_coco/test_client.py   |  10 +-
 .../yolov3_darknet53_270e_coco/test_client.py |  13 +-
 python/paddle_serving_app/reader/__init__.py  |   1 +
 .../paddle_serving_app/reader/image_reader.py | 204 ++++++++++++++++--
 14 files changed, 258 insertions(+), 140 deletions(-)

diff --git a/python/examples/cascade_rcnn/README.md b/python/examples/cascade_rcnn/README.md
index 0f831a400..8029f39a1 100644
--- a/python/examples/cascade_rcnn/README.md
+++ b/python/examples/cascade_rcnn/README.md
@@ -15,7 +15,7 @@ python3 -m paddle_serving_server.serve --model serving_server --port 9292 --gpu_
 
 ### Perform prediction
 ```
-python3 test_client.py 
+python3 test_client.py 000000570688.jpg
 ```
 
 Image with bounding boxes and json result would be saved in `output` folder.
diff --git a/python/examples/cascade_rcnn/README_CN.md b/python/examples/cascade_rcnn/README_CN.md
index 0cc65ed68..828aba8a9 100644
--- a/python/examples/cascade_rcnn/README_CN.md
+++ b/python/examples/cascade_rcnn/README_CN.md
@@ -15,7 +15,7 @@ python3 -m paddle_serving_server.serve --model serving_server --port 9292 --gpu_
 
 ### 执行预测
 ```
-python3 test_client.py
+python3 test_client.py 000000570688.jpg
 ```
 
 客户端已经为图片做好了后处理，在`output`文件夹下存放各个框的json格式信息还有后处理结果图片。
diff --git a/python/examples/cascade_rcnn/test_client.py b/python/examples/cascade_rcnn/test_client.py
index cd90e5f15..aac9f6721 100644
--- a/python/examples/cascade_rcnn/test_client.py
+++ b/python/examples/cascade_rcnn/test_client.py
@@ -18,12 +18,12 @@
 from paddle_serving_app.reader import *
 import cv2
 
-preprocess = Sequential([
-        File2Image(),
-        Resize((800, 1333), True, interpolation=2), 
-        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
-        Transpose((2,0,1)),
-        PadStride(32)
+preprocess = DetectionSequential([
+        DetectionFile2Image(),
+        DetectionResize((800, 1333), True, interpolation=2), 
+        DetectionNormalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        DetectionTranspose((2,0,1)),
+        DetectionPadStride(32)
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
diff --git a/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/README_CN.md b/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/README_CN.md
index 95263591e..11dcbd85f 100644
--- a/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/README_CN.md
+++ b/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/README_CN.md
@@ -10,7 +10,7 @@ wget --no-check-certificate https://paddle-serving.bj.bcebos.com/pddet_demo/fast
 
 ### 启动服务
 ```
-tar xf faster_rcnn_hrnetv2p_w18_1x.tar
+tar xf faster_rcnn_hrnetv2p_w18_1x.tar.gz
 python3 -m paddle_serving_server.serve --model serving_server --port 9494 --gpu_ids 0
 ```
 该模型支持TensorRT，如果想要更快的预测速度，可以开启`--use_trt`选项,但此时需要额外设置子图的TRT变长最大最小最优shape.
diff --git a/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/test_client.py b/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/test_client.py
index b68a0c7b4..329f6effb 100644
--- a/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/test_client.py
+++ b/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x/test_client.py
@@ -18,12 +18,12 @@
 from paddle_serving_app.reader import *
 import cv2
 
-preprocess = Sequential([
-        File2Image(),
-        Resize((800, 1333), True, interpolation=2), 
-        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
-        Transpose((2,0,1)),
-        PadStride(32)
+preprocess = DetectionSequential([
+       DetectionFile2Image(),
+       DetectionResize((800, 1333), True, interpolation=2), 
+       DetectionNormalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+       DetectionTranspose((2,0,1)),
+       DetectionPadStride(32)
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
diff --git a/python/examples/detection/faster_rcnn_r50_fpn_1x_coco/test_client.py b/python/examples/detection/faster_rcnn_r50_fpn_1x_coco/test_client.py
index 5d5e0f2b1..b6b2c534b 100644
--- a/python/examples/detection/faster_rcnn_r50_fpn_1x_coco/test_client.py
+++ b/python/examples/detection/faster_rcnn_r50_fpn_1x_coco/test_client.py
@@ -18,13 +18,13 @@
 from paddle_serving_app.reader import *
 import cv2
 
-preprocess = Sequential([
-        File2Image(),
-        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
-        Resize(
+preprocess = DetectionSequential([
+        DetectionFile2Image(),
+        DetectionNormalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        DetectionResize(
         (800, 1333), True, interpolation=cv2.INTER_LINEAR), 
-        Transpose((2,0,1)),
-        PadStride(128)
+        DetectionTranspose((2,0,1)),
+        DetectionPadStride(128)
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
diff --git a/python/examples/detection/fcos_dcn_r50_fpn_1x_coco/test_client.py b/python/examples/detection/fcos_dcn_r50_fpn_1x_coco/test_client.py
index 4c441f5db..7ad59d75b 100644
--- a/python/examples/detection/fcos_dcn_r50_fpn_1x_coco/test_client.py
+++ b/python/examples/detection/fcos_dcn_r50_fpn_1x_coco/test_client.py
@@ -18,16 +18,16 @@
 from paddle_serving_app.reader import *
 import cv2
 
-preprocess = Sequential([
-        File2Image(),
-        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
-        Resize(
+preprocess = DetectionSequential([
+        DetectionFile2Image(),
+        DetectionNormalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        DetectionResize(
         (800, 1333), True, interpolation=cv2.INTER_LINEAR), 
-        Transpose((2,0,1)),
-        PadStride(128)
+        DetectionTranspose((2,0,1)),
+        DetectionPadStride(128)
 ])
 
-postprocess = RCNNPostprocess("label_list.txt", "output", [608, 608])
+postprocess = RCNNPostprocess("label_list.txt", "output")
 client = Client()
 
 client.load_client_config("serving_client/serving_client_conf.prototxt")
diff --git a/python/examples/detection/ppyolo_r50vd_dcn_1x_coco/test_client.py b/python/examples/detection/ppyolo_r50vd_dcn_1x_coco/test_client.py
index f26e13446..f40f2d5c8 100644
--- a/python/examples/detection/ppyolo_r50vd_dcn_1x_coco/test_client.py
+++ b/python/examples/detection/ppyolo_r50vd_dcn_1x_coco/test_client.py
@@ -18,12 +18,12 @@
 from paddle_serving_app.reader import *
 import cv2
 
-preprocess = Sequential([
-        File2Image(),
-        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
-        Resize(
+preprocess = DetectionSequential([
+        DetectionFile2Image(),
+        DetectionNormalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        DetectionResize(
         (608, 608), False, interpolation=2), 
-        Transpose((2,0,1))
+        DetectionTranspose((2,0,1))
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
diff --git a/python/examples/detection/ssd_vgg16_300_240e_voc/label_list.txt b/python/examples/detection/ssd_vgg16_300_240e_voc/label_list.txt
index 941cb4e13..8420ab35e 100644
--- a/python/examples/detection/ssd_vgg16_300_240e_voc/label_list.txt
+++ b/python/examples/detection/ssd_vgg16_300_240e_voc/label_list.txt
@@ -1,80 +1,20 @@
-person
+aeroplane
 bicycle
-car
-motorcycle
-airplane
-bus
-train
-truck
-boat
-traffic light
-fire hydrant
-stop sign
-parking meter
-bench
 bird
+boat
+bottle
+bus
+car
 cat
+chair
+cow
+diningtable
 dog
 horse
+motorbike
+person
+pottedplant
 sheep
-cow
-elephant
-bear
-zebra
-giraffe
-backpack
-umbrella
-handbag
-tie
-suitcase
-frisbee
-skis
-snowboard
-sports ball
-kite
-baseball bat
-baseball glove
-skateboard
-surfboard
-tennis racket
-bottle
-wine glass
-cup
-fork
-knife
-spoon
-bowl
-banana
-apple
-sandwich
-orange
-broccoli
-carrot
-hot dog
-pizza
-donut
-cake
-chair
-couch
-potted plant
-bed
-dining table
-toilet
-tv
-laptop
-mouse
-remote
-keyboard
-cell phone
-microwave
-oven
-toaster
-sink
-refrigerator
-book
-clock
-vase
-scissors
-teddy bear
-hair drier
-toothbrush
+sofa
+train
+tvmonitor
diff --git a/python/examples/detection/ssd_vgg16_300_240e_voc/test_client.py b/python/examples/detection/ssd_vgg16_300_240e_voc/test_client.py
index 4c441f5db..1df635c89 100644
--- a/python/examples/detection/ssd_vgg16_300_240e_voc/test_client.py
+++ b/python/examples/detection/ssd_vgg16_300_240e_voc/test_client.py
@@ -18,16 +18,15 @@
 from paddle_serving_app.reader import *
 import cv2
 
-preprocess = Sequential([
-        File2Image(),
-        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
-        Resize(
-        (800, 1333), True, interpolation=cv2.INTER_LINEAR), 
-        Transpose((2,0,1)),
-        PadStride(128)
+preprocess = DetectionSequential([
+        DetectionFile2Image(),
+        DetectionResize(
+        (300, 300), False, interpolation=cv2.INTER_LINEAR), 
+        DetectionNormalize([104.0, 117.0, 123.0], [1.0, 1.0, 1.0], False),
+        DetectionTranspose((2,0,1)),
 ])
 
-postprocess = RCNNPostprocess("label_list.txt", "output", [608, 608])
+postprocess = RCNNPostprocess("label_list.txt", "output")
 client = Client()
 
 client.load_client_config("serving_client/serving_client_conf.prototxt")
@@ -37,6 +36,7 @@
 fetch_map = client.predict(
     feed={
         "image": im,
+        "im_shape": np.array(list(im.shape[1:])).reshape(-1),
         "scale_factor": im_info['scale_factor'],
     },
     fetch=["save_infer_model/scale_0.tmp_1"],
diff --git a/python/examples/detection/ttfnet_darknet53_1x_coco/test_client.py b/python/examples/detection/ttfnet_darknet53_1x_coco/test_client.py
index 114f43c1b..f735c01bc 100644
--- a/python/examples/detection/ttfnet_darknet53_1x_coco/test_client.py
+++ b/python/examples/detection/ttfnet_darknet53_1x_coco/test_client.py
@@ -17,12 +17,12 @@
 from paddle_serving_app.reader import *
 import cv2
 
-preprocess = Sequential([
-        File2Image(),
-        Resize(
+preprocess = DetectionSequential([
+        DetectionFile2Image(),
+        DetectionResize(
         (512, 512), False, interpolation=cv2.INTER_LINEAR),
-        Normalize([123.675, 116.28, 103.53], [58.395, 57.12, 57.375], False),
-        Transpose((2,0,1))
+        DetectionNormalize([123.675, 116.28, 103.53], [58.395, 57.12, 57.375], False),
+        DetectionTranspose((2,0,1))
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
diff --git a/python/examples/detection/yolov3_darknet53_270e_coco/test_client.py b/python/examples/detection/yolov3_darknet53_270e_coco/test_client.py
index 5d5e0f2b1..04f21b32a 100644
--- a/python/examples/detection/yolov3_darknet53_270e_coco/test_client.py
+++ b/python/examples/detection/yolov3_darknet53_270e_coco/test_client.py
@@ -18,13 +18,12 @@
 from paddle_serving_app.reader import *
 import cv2
 
-preprocess = Sequential([
-        File2Image(),
-        Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
-        Resize(
-        (800, 1333), True, interpolation=cv2.INTER_LINEAR), 
-        Transpose((2,0,1)),
-        PadStride(128)
+preprocess = DetectionSequential([
+        DetectionFile2Image(),
+        DetectionResize(
+        (608, 608), False, interpolation=2), 
+        DetectionNormalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], True),
+        DetectionTranspose((2,0,1)),
 ])
 
 postprocess = RCNNPostprocess("label_list.txt", "output")
diff --git a/python/paddle_serving_app/reader/__init__.py b/python/paddle_serving_app/reader/__init__.py
index 7d7ebad6f..185b20992 100644
--- a/python/paddle_serving_app/reader/__init__.py
+++ b/python/paddle_serving_app/reader/__init__.py
@@ -13,6 +13,7 @@
 # limitations under the License.
 from .chinese_bert_reader import ChineseBertReader
 from .image_reader import ImageReader, File2Image, URL2Image, Sequential, Normalize, Base64ToImage
+from .image_reader import DetectionFile2Image, DetectionSequential, DetectionNormalize, DetectionTranspose, DetectionResize, DetectionBGR2RGB, DetectionPadStride
 from .image_reader import CenterCrop, Resize, Transpose, Div, RGB2BGR, BGR2RGB, ResizeByFactor
 from .image_reader import RCNNPostprocess, SegPostprocess, PadStride, BlazeFacePostprocess
 from .image_reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
diff --git a/python/paddle_serving_app/reader/image_reader.py b/python/paddle_serving_app/reader/image_reader.py
index 380079c28..24d1759a0 100644
--- a/python/paddle_serving_app/reader/image_reader.py
+++ b/python/paddle_serving_app/reader/image_reader.py
@@ -481,6 +481,37 @@ class Sequential(object):
         >>> ])
     """
 
+    def __init__(self, transforms):
+        self.transforms = transforms
+
+    def __call__(self, img):
+        for t in self.transforms:
+            img = t(img)
+        return img
+
+    def __repr__(self):
+        format_string_ = self.__class__.__name__ + '('
+        for t in self.transforms:
+            format_string_ += '\n'
+            format_string_ += '    {0}'.format(t)
+        format_string_ += '\n)'
+        return format_string_
+
+
+class DetectionSequential(object):
+    """
+    Args:
+        sequence (sequence of ``Transform`` objects): list of transforms to chain.
+
+    This API references some of the design pattern of torchvision
+    Users can simply use this API in training as well
+
+    Example:
+        >>> image_reader.Sequnece([
+        >>>     transforms.CenterCrop(10),
+        >>> ])
+    """
+
     def __init__(self, transforms):
         self.transforms = transforms
 
@@ -518,7 +549,18 @@ class BGR2RGB(object):
     def __init__(self):
         pass
 
-    def __call__(self, img, img_info = False):
+    def __call__(self, img):
+        return img[:, :, ::-1]
+
+    def __repr__(self):
+        return self.__class__.__name__ + "()"
+
+
+class DetectionBGR2RGB(object):
+    def __init__(self):
+        pass
+
+    def __call__(self, img, img_info=None):
         return img[:, :, ::-1], img_info
 
     def __repr__(self):
@@ -542,7 +584,30 @@ class File2Image(object):
     def __init__(self):
         pass
 
-    def __call__(self, img_path, im_info):
+    def __call__(self, img_path):
+        if py_version == 2:
+            fin = open(img_path)
+        else:
+            fin = open(img_path, "rb")
+        sample = fin.read()
+        data = np.fromstring(sample, np.uint8)
+        img = cv2.imdecode(data, cv2.IMREAD_COLOR)
+        '''
+        img = cv2.imread(img_path, -1)
+        channels = img.shape[2]
+        ori_h = img.shape[0]
+        ori_w = img.shape[1]
+        '''
+        return img
+
+    def __repr__(self):
+        return self.__class__.__name__ + "()"
+
+class DetectionFile2Image(object):
+    def __init__(self):
+        pass
+
+    def __call__(self, img_path, im_info=None):
         if py_version == 2:
             fin = open(img_path)
         else:
@@ -557,8 +622,9 @@ def __call__(self, img_path, im_info):
         ori_h = img.shape[0]
         ori_w = img.shape[1]
         '''
-        im_info['im_shape'] = np.array(img.shape[:2], dtype=np.float32)
-        im_info['scale_factor'] = np.array([1., 1.], dtype=np.float32)
+        if im_info is not None:
+            im_info['im_shape'] = np.array(img.shape[:2], dtype=np.float32)
+            im_info['scale_factor'] = np.array([1., 1.], dtype=np.float32)
         return img, im_info
 
     def __repr__(self):
@@ -600,7 +666,28 @@ class Div(object):
     def __init__(self, value):
         self.value = value
 
-    def __call__(self, img, img_info = False):
+    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 DetectionDiv(object):
+    """ divide by some float number """
+
+    def __init__(self, value):
+        self.value = value
+
+    def __call__(self, img, img_info=None):
         """
         Args:
             img (numpy array): (int8 numpy array)
@@ -622,6 +709,41 @@ class Normalize(object):
     will normalize each channel of the input ``torch.*Tensor`` i.e.
     ``output[channel] = (input[channel] - mean[channel]) / std[channel]``
 
+    .. note::
+        This transform acts out of place, i.e., it does not mutate the input tensor.
+
+    Args:
+        mean (sequence): Sequence of means for each channel.
+        std (sequence): Sequence of standard deviations for each channel.
+
+    """
+
+    def __init__(self, mean, std, channel_first=False):
+        self.mean = mean
+        self.std = std
+        self.channel_first = channel_first
+
+    def __call__(self, img):
+        """
+        Args:
+            img (numpy array): (C, H, W) to be normalized.
+
+        Returns:
+            Tensor: Normalized Tensor image.
+        """
+        return F.normalize(img, self.mean, self.std, self.channel_first)
+
+    def __repr__(self):
+        return self.__class__.__name__ + '(mean={0}, std={1})'.format(self.mean,
+                                                                      self.std)
+
+
+class DetectionNormalize(object):
+    """Normalize a tensor image with mean and standard deviation.
+    Given mean: ``(M1,...,Mn)`` and std: ``(S1,..,Sn)`` for ``n`` channels, this transform
+    will normalize each channel of the input ``torch.*Tensor`` i.e.
+    ``output[channel] = (input[channel] - mean[channel]) / std[channel]``
+
     .. note::
         This transform acts out of place, i.e., it does not mutate the input tensor.
 
@@ -637,7 +759,7 @@ def __init__(self, mean, std, is_scale=True):
         self.std = std
         self.is_scale = is_scale
 
-    def __call__(self, im, im_info = None):
+    def __call__(self, im, im_info=None):
         """
         Args:
             im (np.ndarray): image (np.ndarray)
@@ -708,8 +830,33 @@ def __repr__(self):
         return self.__class__.__name__ + '(size={0})'.format(self.size)
 
 
-
 class Resize(object):
+    """Resize the input numpy array Image to the given size.
+
+    Args:
+        size (sequence or int): Desired output size. If size is a sequence like
+            (w, h), output size will be matched to this. If size is an int,
+            smaller edge of the image will be matched to this number.
+            i.e, if height > width, then image will be rescaled to
+            (size * height / width, size)
+        interpolation (int, optional): Desired interpolation. Default is
+            ``None``
+    """
+
+    def __init__(self, size, max_size=2147483647, interpolation=None):
+        self.size = size
+        self.max_size = max_size
+        self.interpolation = interpolation
+
+    def __call__(self, img):
+        return F.resize(img, self.size, self.max_size, self.interpolation)
+
+    def __repr__(self):
+        return self.__class__.__name__ + '(size={0}, max_size={1}, interpolation={2})'.format(
+            self.size, self.max_size,
+            _cv2_interpolation_to_str[self.interpolation])
+
+class DetectionResize(object):
     """resize image by target_size and max_size
     Args:
         target_size (int): the target size of image
@@ -724,7 +871,7 @@ def __init__(self, target_size, keep_ratio=True, interpolation=cv2.INTER_LINEAR)
         self.keep_ratio = keep_ratio
         self.interpolation = interpolation
 
-    def __call__(self, im, im_info):
+    def __call__(self, im, im_info=None):
         """
         Args:
             im (np.ndarray): image (np.ndarray)
@@ -744,9 +891,10 @@ def __call__(self, im, im_info):
             fx=im_scale_x,
             fy=im_scale_y,
             interpolation=self.interpolation)
-        im_info['im_shape'] = np.array(im.shape[:2]).astype('float32')
-        im_info['scale_factor'] = np.array(
-            [im_scale_y, im_scale_x]).astype('float32')
+        if im_info is not None:
+            im_info['im_shape'] = np.array(im.shape[:2]).astype('float32')
+            im_info['scale_factor'] = np.array(
+                [im_scale_y, im_scale_x]).astype('float32')
         return im, im_info
 
     def generate_scale(self, im):
@@ -782,6 +930,23 @@ def __repr__(self):
 
 
 class PadStride(object):
+    def __init__(self, stride):
+        self.coarsest_stride = stride
+
+    def __call__(self, img):
+        coarsest_stride = self.coarsest_stride
+        if coarsest_stride == 0:
+            return img
+        im_c, im_h, im_w = img.shape
+        pad_h = int(np.ceil(float(im_h) / coarsest_stride) * coarsest_stride)
+        pad_w = int(np.ceil(float(im_w) / coarsest_stride) * coarsest_stride)
+        padding_im = np.zeros((im_c, pad_h, pad_w), dtype=np.float32)
+        padding_im[:, :im_h, :im_w] = img
+        im_info = {}
+        im_info['resize_shape'] = padding_im.shape[1:]
+        return padding_im
+
+class DetectionPadStride(object):
     """ padding image for model with FPN, instead PadBatch(pad_to_stride) in original config
     Args:
         stride (bool): model with FPN need image shape % stride == 0
@@ -790,7 +955,7 @@ class PadStride(object):
     def __init__(self, stride=0):
         self.coarsest_stride = stride
 
-    def __call__(self, im, im_info = None):
+    def __call__(self, im, im_info=None):
         """
         Args:
             im (np.ndarray): image (np.ndarray)
@@ -865,7 +1030,20 @@ class Transpose(object):
     def __init__(self, transpose_target):
         self.transpose_target = transpose_target
 
-    def __call__(self, im, im_info = None):
+    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 DetectionTranspose(object):
+    def __init__(self, transpose_target):
+        self.transpose_target = transpose_target
+
+    def __call__(self, im, im_info=None):
         im = F.transpose(im, self.transpose_target)
         return im, im_info
 

From 60ad0d35a682774ae23a508c00cf2ff326b43cc5 Mon Sep 17 00:00:00 2001
From: huangjianhui <852142024@qq.com>
Date: Tue, 19 Oct 2021 12:14:46 +0800
Subject: [PATCH 7/7] Update test_client.py

---
 python/examples/yolov4/test_client.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/examples/yolov4/test_client.py b/python/examples/yolov4/test_client.py
index 3db9b3ec3..dfcd58610 100644
--- a/python/examples/yolov4/test_client.py
+++ b/python/examples/yolov4/test_client.py
@@ -29,7 +29,7 @@
 client.load_client_config("yolov4_client/serving_client_conf.prototxt")
 client.connect(['127.0.0.1:9393'])
 
-im, im_info = preprocess(sys.argv[1])
+im = preprocess(sys.argv[1])
 fetch_map = client.predict(
     feed={
         "image": im,