diff --git a/lite/CMakeLists.txt b/lite/CMakeLists.txt
index cb6a872e061a51f142bd2301171f0559a1ccb129..bac6f80c4721e0c5de201eebfe7e6a39a0bdc73a 100644
--- a/lite/CMakeLists.txt
+++ b/lite/CMakeLists.txt
@@ -232,6 +232,8 @@ if (LITE_WITH_LIGHT_WEIGHT_FRAMEWORK AND LITE_WITH_ARM)
                 COMMAND cp "${CMAKE_SOURCE_DIR}/lite/demo/cxx/makefiles/mobile_classify/Makefile.${ARM_TARGET_OS}.${ARM_TARGET_ARCH_ABI}" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx/mobile_classify/Makefile"
                 COMMAND cp -r "${CMAKE_SOURCE_DIR}/lite/demo/cxx/test_cv" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx"
                 COMMAND cp "${CMAKE_SOURCE_DIR}/lite/demo/cxx/makefiles/test_cv/Makefile.${ARM_TARGET_OS}.${ARM_TARGET_ARCH_ABI}" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx/test_cv/Makefile"
+                COMMAND cp -r "${CMAKE_SOURCE_DIR}/lite/demo/cxx/mask_detection" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx"
+                COMMAND cp "${CMAKE_SOURCE_DIR}/lite/demo/cxx/makefiles/mask_detection/Makefile.${ARM_TARGET_OS}.${ARM_TARGET_ARCH_ABI}" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx/mask_detection/Makefile"
             )
             add_dependencies(publish_inference_android_cxx_demos logging gflags)
             add_dependencies(publish_inference_cxx_lib publish_inference_android_cxx_demos)
@@ -251,6 +253,8 @@ if (LITE_WITH_LIGHT_WEIGHT_FRAMEWORK AND LITE_WITH_ARM)
                 COMMAND cp "${CMAKE_SOURCE_DIR}/lite/demo/cxx/makefiles/mobile_classify/Makefile.${ARM_TARGET_OS}.${ARM_TARGET_ARCH_ABI}" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx/mobile_classify/Makefile"
                 COMMAND cp -r "${CMAKE_SOURCE_DIR}/lite/demo/cxx/test_cv" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx"
                 COMMAND cp "${CMAKE_SOURCE_DIR}/lite/demo/cxx/makefiles/test_cv/Makefile.${ARM_TARGET_OS}.${ARM_TARGET_ARCH_ABI}" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx/test_cv/Makefile"
+                COMMAND cp -r "${CMAKE_SOURCE_DIR}/lite/demo/cxx/mask_detection" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx"
+                COMMAND cp "${CMAKE_SOURCE_DIR}/lite/demo/cxx/makefiles/mask_detection/Makefile.${ARM_TARGET_OS}.${ARM_TARGET_ARCH_ABI}" "${INFER_LITE_PUBLISH_ROOT}/demo/cxx/mask_detection/Makefile"
             )
             add_dependencies(tiny_publish_cxx_lib publish_inference_android_cxx_demos)
         endif()
diff --git a/lite/demo/cxx/README.md b/lite/demo/cxx/README.md
index 3217a7ed49006325715e22f8aa82d155bc8bf927..7760d5df2b82887b331cb4a536fe9e89a95c0ce7 100644
--- a/lite/demo/cxx/README.md
+++ b/lite/demo/cxx/README.md
@@ -71,25 +71,25 @@ tar zxvf mobilenet_v1.tar.gz
 ./model_optimize_tool optimize model
 make
 
-adb -s emulator-5554 push mobile_classify /data/local/tmp/
-adb -s emulator-5554 push test.jpg /data/local/tmp/
-adb -s emulator-5554 push labels.txt /data/local/tmp/
-adb -s emulator-5554 push ../../../cxx/lib/libpaddle_light_api_shared.so /data/local/tmp/
-adb -s emulator-5554 shell chmod +x /data/local/tmp/mobile_classify
-adb -s emulator-5554 shell "export LD_LIBRARY_PATH=/data/local/tmp/:$LD_LIBRARY_PATH && 
+adb push mobile_classify /data/local/tmp/
+adb push test.jpg /data/local/tmp/
+adb push labels.txt /data/local/tmp/
+adb push ../../../cxx/lib/libpaddle_light_api_shared.so /data/local/tmp/
+adb shell chmod +x /data/local/tmp/mobile_classify
+adb shell "export LD_LIBRARY_PATH=/data/local/tmp/:$LD_LIBRARY_PATH && 
 /data/local/tmp/mobile_classify /data/local/tmp/mobilenetv1opt2 /data/local/tmp/test.jpg /data/local/tmp/labels.txt"
 ```
 运行成功将在控制台输出预测结果的前5个类别的预测概率
 - 如若想看前10个类别的预测概率，在运行命令输入topk的值即可
     eg:
     ```shell
-    adb -s emulator-5554 shell "export LD_LIBRARY_PATH=/data/local/tmp/:$LD_LIBRARY_PATH && 
+    adb shell "export LD_LIBRARY_PATH=/data/local/tmp/:$LD_LIBRARY_PATH && 
     /data/local/tmp/mobile_classify /data/local/tmp/mobilenetv1opt2/ /data/local/tmp/test.jpg /data/local/tmp/labels.txt 10"
     ```
 - 如若想看其他模型的分类结果， 在运行命令输入model_dir 及其model的输入大小即可
     eg:
     ```shell
-    adb -s emulator-5554 shell "export LD_LIBRARY_PATH=/data/local/tmp/:$LD_LIBRARY_PATH && 
+    adb shell "export LD_LIBRARY_PATH=/data/local/tmp/:$LD_LIBRARY_PATH && 
     /data/local/tmp/mobile_classify /data/local/tmp/mobilenetv2opt2/ /data/local/tmp/test.jpg /data/local/tmp/labels.txt 10 224 224"
     ```
     
@@ -100,12 +100,34 @@ wget http://paddle-inference-dist.bj.bcebos.com/mobilenet_v1.tar.gz
 tar zxvf mobilenet_v1.tar.gz
 ./model_optimize_tool optimize model
 make
-adb -s emulator-5554 push test_model_cv /data/local/tmp/
-adb -s emulator-5554 push test.jpg /data/local/tmp/
-adb -s emulator-5554 push labels.txt /data/local/tmp/
-adb -s emulator-5554 push ../../../cxx/lib/libpaddle_full_api_shared.so /data/local/tmp/
-adb -s emulator-5554 shell chmod +x /data/local/tmp/test_model_cv
-adb -s emulator-5554 shell "export LD_LIBRARY_PATH=/data/local/tmp/:$LD_LIBRARY_PATH && 
+adb push test_model_cv /data/local/tmp/
+adb push test.jpg /data/local/tmp/
+adb push labels.txt /data/local/tmp/
+adb push ../../../cxx/lib/libpaddle_full_api_shared.so /data/local/tmp/
+adb shell chmod +x /data/local/tmp/test_model_cv
+adb shell "export LD_LIBRARY_PATH=/data/local/tmp/:$LD_LIBRARY_PATH && 
 /data/local/tmp/test_model_cv /data/local/tmp/mobilenetv1opt2 /data/local/tmp/test.jpg /data/local/tmp/labels.txt"
 ```
 运行成功将在控制台输出预测结果的前10个类别的预测概率
+
+10. 编译并运行mask_detection口罩检测的demo
+
+注：运行该demo所需的libpaddle_light_api_shared.so，编译选项需使用build_extra=ON
+
+```shell
+cd ../mask_detection
+wget https://paddle-inference-dist.bj.bcebos.com/mask_detection.tar.gz
+tar zxvf mask_detection.tar.gz
+make
+adb push mask_detection /data/local/tmp/
+adb push test.jpg /data/local/tmp/
+adb push face_detection /data/local/tmp
+adb push mask_classification /data/local/tmp
+adb push ../../../cxx/lib/libpaddle_light_api_shared.so /data/local/tmp/
+adb shell chmod +x /data/local/tmp/mask_detection
+adb shell "export LD_LIBRARY_PATH=/data/local/tmp/:$LD_LIBRARY_PATH && 
+/data/local/tmp/mask_detection /data/local/tmp/face_detection \
+/data/local/tmp/mask_classification /data/local/tmp/test.jpg"
+adb pull /data/local/tmp/test_mask_detection_result.jpg ./
+```
+运行成功将在mask_detection目录下看到生成的口罩检测结果图像: test_mask_detection_result.jpg
diff --git a/lite/demo/cxx/makefiles/mask_detection/Makefile.android.armv7 b/lite/demo/cxx/makefiles/mask_detection/Makefile.android.armv7
new file mode 100644
index 0000000000000000000000000000000000000000..dd6d4b0960160e140e2f051b78814d2fee08d5e0
--- /dev/null
+++ b/lite/demo/cxx/makefiles/mask_detection/Makefile.android.armv7
@@ -0,0 +1,61 @@
+ARM_ABI = arm7
+export ARM_ABI
+
+include ../Makefile.def
+
+LITE_ROOT=../../../
+
+THIRD_PARTY_DIR=${LITE_ROOT}/third_party
+
+OPENCV_VERSION=opencv4.1.0
+
+OPENCV_LIBS = ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/libs/libopencv_imgcodecs.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/libs/libopencv_imgproc.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/libs/libopencv_core.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/3rdparty/libs/libtegra_hal.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/3rdparty/libs/liblibjpeg-turbo.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/3rdparty/libs/liblibwebp.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/3rdparty/libs/liblibpng.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/3rdparty/libs/liblibjasper.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/3rdparty/libs/liblibtiff.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/3rdparty/libs/libIlmImf.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/3rdparty/libs/libtbb.a \
+              ../../../third_party/${OPENCV_VERSION}/armeabi-v7a/3rdparty/libs/libcpufeatures.a
+
+OPENCV_INCLUDE = -I../../../third_party/${OPENCV_VERSION}/armeabi-v7a/include
+
+CXX_INCLUDES = $(INCLUDES) ${OPENCV_INCLUDE} -I$(LITE_ROOT)/cxx/include
+
+CXX_LIBS = ${OPENCV_LIBS} -L$(LITE_ROOT)/cxx/lib/ -lpaddle_light_api_shared $(SYSTEM_LIBS)
+
+###############################################################
+# How to use one of static libaray:                           #
+#  `libpaddle_api_full_bundled.a`                             #
+#  `libpaddle_api_light_bundled.a`                            #
+###############################################################
+# Note: default use lite's shared library.                    #
+###############################################################
+# 1. Comment above line using `libpaddle_light_api_shared.so`
+# 2. Undo comment below line using `libpaddle_api_light_bundled.a`
+
+#CXX_LIBS = $(LITE_ROOT)/cxx/lib/libpaddle_api_light_bundled.a $(SYSTEM_LIBS)
+
+mask_detection: fetch_opencv mask_detection.o
+	$(CC) $(SYSROOT_LINK) $(CXXFLAGS_LINK) mask_detection.o -o mask_detection  $(CXX_LIBS) $(LDFLAGS)
+
+mask_detection.o: mask_detection.cc
+	$(CC) $(SYSROOT_COMPLILE) $(CXX_DEFINES) $(CXX_INCLUDES) $(CXX_FLAGS) -o mask_detection.o -c mask_detection.cc
+
+fetch_opencv:
+	@ test -d ${THIRD_PARTY_DIR} ||  mkdir ${THIRD_PARTY_DIR}
+	@ test -e ${THIRD_PARTY_DIR}/${OPENCV_VERSION}.tar.gz || \
+      (echo "fetch opencv libs" && \
+      wget -P ${THIRD_PARTY_DIR} https://paddle-inference-dist.bj.bcebos.com/${OPENCV_VERSION}.tar.gz)
+	@ test -d ${THIRD_PARTY_DIR}/${OPENCV_VERSION} || \
+      tar -zxvf ${THIRD_PARTY_DIR}/${OPENCV_VERSION}.tar.gz -C ${THIRD_PARTY_DIR}
+
+
+.PHONY: clean
+clean:
+	rm -f mask_detection.o
+	rm -f mask_detection
diff --git a/lite/demo/cxx/makefiles/mask_detection/Makefile.android.armv8 b/lite/demo/cxx/makefiles/mask_detection/Makefile.android.armv8
new file mode 100644
index 0000000000000000000000000000000000000000..c2f601ed2f68c342b47c5add451f84c537f978de
--- /dev/null
+++ b/lite/demo/cxx/makefiles/mask_detection/Makefile.android.armv8
@@ -0,0 +1,61 @@
+ARM_ABI = arm8
+export ARM_ABI
+
+include ../Makefile.def
+
+LITE_ROOT=../../../
+
+THIRD_PARTY_DIR=${LITE_ROOT}/third_party
+
+OPENCV_VERSION=opencv4.1.0
+
+OPENCV_LIBS = ../../../third_party/${OPENCV_VERSION}/arm64-v8a/libs/libopencv_imgcodecs.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/libs/libopencv_imgproc.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/libs/libopencv_core.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/libtegra_hal.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibjpeg-turbo.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibwebp.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibpng.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibjasper.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibtiff.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/libIlmImf.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/libtbb.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/libcpufeatures.a
+
+OPENCV_INCLUDE = -I../../../third_party/${OPENCV_VERSION}/arm64-v8a/include
+
+CXX_INCLUDES = $(INCLUDES) ${OPENCV_INCLUDE} -I$(LITE_ROOT)/cxx/include
+
+CXX_LIBS = ${OPENCV_LIBS} -L$(LITE_ROOT)/cxx/lib/ -lpaddle_light_api_shared $(SYSTEM_LIBS)
+
+###############################################################
+# How to use one of static libaray:                           #
+#  `libpaddle_api_full_bundled.a`                             #
+#  `libpaddle_api_light_bundled.a`                            #
+###############################################################
+# Note: default use lite's shared library.                    #
+###############################################################
+# 1. Comment above line using `libpaddle_light_api_shared.so`
+# 2. Undo comment below line using `libpaddle_api_light_bundled.a`
+
+#CXX_LIBS = $(LITE_ROOT)/cxx/lib/libpaddle_api_light_bundled.a $(SYSTEM_LIBS)
+
+mask_detection: fetch_opencv mask_detection.o
+	$(CC) $(SYSROOT_LINK) $(CXXFLAGS_LINK) mask_detection.o -o mask_detection  $(CXX_LIBS) $(LDFLAGS)
+
+mask_detection.o: mask_detection.cc
+	$(CC) $(SYSROOT_COMPLILE) $(CXX_DEFINES) $(CXX_INCLUDES) $(CXX_FLAGS) -o mask_detection.o -c mask_detection.cc
+
+fetch_opencv:
+	@ test -d ${THIRD_PARTY_DIR} ||  mkdir ${THIRD_PARTY_DIR}
+	@ test -e ${THIRD_PARTY_DIR}/${OPENCV_VERSION}.tar.gz || \
+      (echo "fetch opencv libs" && \
+      wget -P ${THIRD_PARTY_DIR} https://paddle-inference-dist.bj.bcebos.com/${OPENCV_VERSION}.tar.gz)
+	@ test -d ${THIRD_PARTY_DIR}/${OPENCV_VERSION} || \
+      tar -zxvf ${THIRD_PARTY_DIR}/${OPENCV_VERSION}.tar.gz -C ${THIRD_PARTY_DIR}
+
+
+.PHONY: clean
+clean:
+	rm -f mask_detection.o
+	rm -f mask_detection
diff --git a/lite/demo/cxx/mask_detection/mask_detection.cc b/lite/demo/cxx/mask_detection/mask_detection.cc
new file mode 100644
index 0000000000000000000000000000000000000000..748b84365fc70aa59171a6bf8847f554308fdc8c
--- /dev/null
+++ b/lite/demo/cxx/mask_detection/mask_detection.cc
@@ -0,0 +1,246 @@
+// Copyright (c) 2019 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.
+
+#include <iostream>
+#include <string>
+#include <vector>
+#include "opencv2/core.hpp"
+#include "opencv2/imgcodecs.hpp"
+#include "opencv2/imgproc.hpp"
+#include "paddle_api.h"  // NOLINT
+
+using namespace paddle::lite_api;  // NOLINT
+
+struct Object {
+  int batch_id;
+  cv::Rect rec;
+  int class_id;
+  float prob;
+};
+
+int64_t ShapeProduction(const shape_t& shape) {
+  int64_t res = 1;
+  for (auto i : shape) res *= i;
+  return res;
+}
+
+// fill tensor with mean and scale and trans layout: nhwc -> nchw, neon speed up
+void neon_mean_scale(const float* din,
+                     float* dout,
+                     int size,
+                     const std::vector<float> mean,
+                     const std::vector<float> scale) {
+  if (mean.size() != 3 || scale.size() != 3) {
+    std::cerr << "[ERROR] mean or scale size must equal to 3\n";
+    exit(1);
+  }
+  float32x4_t vmean0 = vdupq_n_f32(mean[0]);
+  float32x4_t vmean1 = vdupq_n_f32(mean[1]);
+  float32x4_t vmean2 = vdupq_n_f32(mean[2]);
+  float32x4_t vscale0 = vdupq_n_f32(scale[0]);
+  float32x4_t vscale1 = vdupq_n_f32(scale[1]);
+  float32x4_t vscale2 = vdupq_n_f32(scale[2]);
+
+  float* dout_c0 = dout;
+  float* dout_c1 = dout + size;
+  float* dout_c2 = dout + size * 2;
+
+  int i = 0;
+  for (; i < size - 3; i += 4) {
+    float32x4x3_t vin3 = vld3q_f32(din);
+    float32x4_t vsub0 = vsubq_f32(vin3.val[0], vmean0);
+    float32x4_t vsub1 = vsubq_f32(vin3.val[1], vmean1);
+    float32x4_t vsub2 = vsubq_f32(vin3.val[2], vmean2);
+    float32x4_t vs0 = vmulq_f32(vsub0, vscale0);
+    float32x4_t vs1 = vmulq_f32(vsub1, vscale1);
+    float32x4_t vs2 = vmulq_f32(vsub2, vscale2);
+    vst1q_f32(dout_c0, vs0);
+    vst1q_f32(dout_c1, vs1);
+    vst1q_f32(dout_c2, vs2);
+
+    din += 12;
+    dout_c0 += 4;
+    dout_c1 += 4;
+    dout_c2 += 4;
+  }
+  for (; i < size; i++) {
+    *(dout_c0++) = (*(din++) - mean[0]) * scale[0];
+    *(dout_c1++) = (*(din++) - mean[1]) * scale[1];
+    *(dout_c2++) = (*(din++) - mean[2]) * scale[2];
+  }
+}
+
+void pre_process(const cv::Mat& img,
+                 int width,
+                 int height,
+                 const std::vector<float>& mean,
+                 const std::vector<float>& scale,
+                 float* data,
+                 bool is_scale = false) {
+  cv::Mat resized_img;
+  cv::resize(
+      img, resized_img, cv::Size(width, height), 0.f, 0.f, cv::INTER_CUBIC);
+  cv::Mat imgf;
+  float scale_factor = is_scale ? 1.f / 256 : 1.f;
+  resized_img.convertTo(imgf, CV_32FC3, scale_factor);
+  const float* dimg = reinterpret_cast<const float*>(imgf.data);
+  neon_mean_scale(dimg, data, width * height, mean, scale);
+}
+
+void RunModel(std::string det_model_dir,
+              std::string class_model_dir,
+              std::string img_path) {
+  // Prepare
+  cv::Mat img = imread(img_path, cv::IMREAD_COLOR);
+  float shrink = 0.2;
+  int width = img.cols;
+  int height = img.rows;
+  int s_width = static_cast<int>(width * shrink);
+  int s_height = static_cast<int>(height * shrink);
+
+  // Detection
+  MobileConfig config;
+  config.set_model_dir(det_model_dir);
+
+  // Create Predictor For Detction Model
+  std::shared_ptr<PaddlePredictor> predictor =
+      CreatePaddlePredictor<MobileConfig>(config);
+
+  // Get Input Tensor
+  std::unique_ptr<Tensor> input_tensor0(std::move(predictor->GetInput(0)));
+  input_tensor0->Resize({1, 3, s_height, s_width});
+  auto* data = input_tensor0->mutable_data<float>();
+
+  // Do PreProcess
+  std::vector<float> detect_mean = {104.f, 117.f, 123.f};
+  std::vector<float> detect_scale = {0.007843, 0.007843, 0.007843};
+  pre_process(img, s_width, s_height, detect_mean, detect_scale, data, false);
+
+  // Detection Model Run
+  predictor->Run();
+
+  // Get Output Tensor
+  std::unique_ptr<const Tensor> output_tensor0(
+      std::move(predictor->GetOutput(0)));
+  auto* outptr = output_tensor0->data<float>();
+  auto shape_out = output_tensor0->shape();
+  int64_t out_len = ShapeProduction(shape_out);
+
+  // Filter Out Detection Box
+  float detect_threshold = 0.3;
+  std::vector<Object> detect_result;
+  for (int i = 0; i < out_len / 6; ++i) {
+    if (outptr[1] >= detect_threshold) {
+      Object obj;
+      int xmin = static_cast<int>(width * outptr[2]);
+      int ymin = static_cast<int>(height * outptr[3]);
+      int xmax = static_cast<int>(width * outptr[4]);
+      int ymax = static_cast<int>(height * outptr[5]);
+      int w = xmax - xmin;
+      int h = ymax - ymin;
+      cv::Rect rec_clip =
+          cv::Rect(xmin, ymin, w, h) & cv::Rect(0, 0, width, height);
+      obj.rec = rec_clip;
+      detect_result.push_back(obj);
+    }
+    outptr += 6;
+  }
+
+  // Classification
+  config.set_model_dir(class_model_dir);
+
+  // Create Predictor For Classification Model
+  predictor = CreatePaddlePredictor<MobileConfig>(config);
+
+  // Get Input Tensor
+  std::unique_ptr<Tensor> input_tensor1(std::move(predictor->GetInput(0)));
+  int classify_w = 128;
+  int classify_h = 128;
+  input_tensor1->Resize({1, 3, classify_h, classify_w});
+  auto* input_data = input_tensor1->mutable_data<float>();
+  int detect_num = detect_result.size();
+  std::vector<float> classify_mean = {0.5f, 0.5f, 0.5f};
+  std::vector<float> classify_scale = {1.f, 1.f, 1.f};
+  float classify_threshold = 0.5;
+  for (int i = 0; i < detect_num; ++i) {
+    cv::Rect rec_clip = detect_result[i].rec;
+    cv::Mat roi = img(rec_clip);
+
+    // Do PreProcess
+    pre_process(roi,
+                classify_w,
+                classify_h,
+                classify_mean,
+                classify_scale,
+                input_data,
+                true);
+
+    // Classification Model Run
+    predictor->Run();
+
+    // Get Output Tensor
+    std::unique_ptr<const Tensor> output_tensor1(
+        std::move(predictor->GetOutput(1)));
+    auto* outptr = output_tensor1->data<float>();
+
+    // Draw Detection and Classification Results
+    cv::rectangle(img, rec_clip, cv::Scalar(0, 0, 255), 2, cv::LINE_AA);
+    std::string text = outptr[1] > classify_threshold ? "wear mask" : "no mask";
+    int font_face = cv::FONT_HERSHEY_COMPLEX_SMALL;
+    double font_scale = 1.f;
+    int thickness = 1;
+    cv::Size text_size =
+        cv::getTextSize(text, font_face, font_scale, thickness, nullptr);
+    float new_font_scale = rec_clip.width * 0.7 * font_scale / text_size.width;
+    text_size =
+        cv::getTextSize(text, font_face, new_font_scale, thickness, nullptr);
+    cv::Point origin;
+    origin.x = rec_clip.x + 5;
+    origin.y = rec_clip.y + text_size.height + 5;
+    cv::putText(img,
+                text,
+                origin,
+                font_face,
+                new_font_scale,
+                cv::Scalar(0, 255, 255),
+                thickness,
+                cv::LINE_AA);
+
+    std::cout << "detect face, location: x=" << rec_clip.x
+              << ", y=" << rec_clip.y << ", width=" << rec_clip.width
+              << ", height=" << rec_clip.height
+              << ", wear mask: " << (outptr[1] > classify_threshold)
+              << std::endl;
+  }
+
+  // Write Result to Image File
+  int start = img_path.find_last_of("/");
+  int end = img_path.find_last_of(".");
+  std::string img_name = img_path.substr(start + 1, end - start - 1);
+  std::string result_name = img_name + "_mask_detection_result.jpg";
+  cv::imwrite(result_name, img);
+}
+
+int main(int argc, char** argv) {
+  if (argc < 3) {
+    std::cerr << "[ERROR] usage: " << argv[0]
+              << " detction_model_dir classification_model_dir image_path\n";
+    exit(1);
+  }
+  std::string detect_model_dir = argv[1];
+  std::string classify_model_dir = argv[2];
+  std::string img_path = argv[3];
+  RunModel(detect_model_dir, classify_model_dir, img_path);
+  return 0;
+}
diff --git a/lite/demo/cxx/ssd_detection/ssd_detection.cc b/lite/demo/cxx/ssd_detection/ssd_detection.cc
index 011733eb87f551141c52ab8e23d9625c93c742fc..2408afcbf64a24924eca119a9d9481dc030250c9 100644
--- a/lite/demo/cxx/ssd_detection/ssd_detection.cc
+++ b/lite/demo/cxx/ssd_detection/ssd_detection.cc
@@ -82,8 +82,8 @@ void neon_mean_scale(const float* din,
   }
   for (; i < size; i++) {
     *(dout_c0++) = (*(din++) - mean[0]) * scale[0];
-    *(dout_c0++) = (*(din++) - mean[1]) * scale[1];
-    *(dout_c0++) = (*(din++) - mean[2]) * scale[2];
+    *(dout_c1++) = (*(din++) - mean[1]) * scale[1];
+    *(dout_c2++) = (*(din++) - mean[2]) * scale[2];
   }
 }
 
@@ -188,13 +188,12 @@ void RunModel(std::string model_dir, std::string img_path) {
       std::move(predictor->GetOutput(0)));
   auto* outptr = output_tensor->data<float>();
   auto shape_out = output_tensor->shape();
-  int64_t cnt = 1;
-  for (auto& i : shape_out) {
-    cnt *= i;
-  }
+  int64_t cnt = ShapeProduction(shape_out);
   auto rec_out = detect_object(outptr, static_cast<int>(cnt / 6), 0.6f, img);
-  std::string result_name =
-      img_path.substr(0, img_path.find(".")) + "_ssd_detection_result.jpg";
+  int start = img_path.find_last_of("/");
+  int end = img_path.find_last_of(".");
+  std::string img_name = img_path.substr(start + 1, end - start - 1);
+  std::string result_name = img_name + "_ssd_detection_result.jpg";
   cv::imwrite(result_name, img);
 }