Merge pull request #577 from wangjiawei04/cascade

add cascade rcnn

python/examples/cascade_rcnn/README.md

+# Cascade RCNN model on Paddle Serving
+
+([简体中文](./README_CN.md)|English)
+
+### Get The Cascade RCNN Model
+```
+sh get_data.sh
+```
+If you want to have more detection models, please refer to [Paddle Detection Model Zoo](https://github.com/PaddlePaddle/PaddleDetection/blob/release/0.2/docs/MODEL_ZOO_cn.md)
+
+### Start the service
+```
+python -m paddle_serving_server_gpu.serve --model serving_server --port 9292 --gpu_id 0
+```
+
+### Perform prediction
+```
+python test_client.py 
+```
+
+Image with bounding boxes and json result would be saved in `output` folder.

python/examples/cascade_rcnn/README_CN.md

+# 使用Paddle Serving部署Cascade RCNN模型
+
+(简体中文|[English](./README.md))
+
+## 获得Cascade RCNN模型
+```
+sh get_data.sh
+```
+如果你想要更多的检测模型，请参考[Paddle检测模型库](https://github.com/PaddlePaddle/PaddleDetection/blob/release/0.2/docs/MODEL_ZOO_cn.md)
+
+### 启动服务
+```
+python -m paddle_serving_server_gpu.serve --model serving_server --port 9292 --gpu_id 0
+```
+
+### 执行预测
+```
+python test_client.py
+```
+
+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。

python/examples/cascade_rcnn/get_data.sh

+wget --no-check-certificate https://paddle-serving.bj.bcebos.com/pddet_demo/cascade_rcnn_r50_fpx_1x_serving.tar.gz
+tar xf cascade_rcnn_r50_fpx_1x_serving.tar.gz

python/paddle_serving_app/reader/image_reader.py

@@ -465,6 +465,24 @@ class Resize(object):
             _cv2_interpolation_to_str[self.interpolation])
 
 
+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