Merge pull request #5576 from WenmuZhou/android

[android demo] Fix error when run same image

deploy/android_demo/README.md

-# 如何快速测试
-### 1. 安装最新版本的Android Studio
+- [Android Demo](#android-demo)
+  - [1. 简介](#1-简介)
+  - [2. 近期更新](#2-近期更新)
+  - [3. 快速使用](#3-快速使用)
+    - [3.1 安装最新版本的Android Studio](#31-安装最新版本的android-studio)
+    - [3.2 安装 NDK 20 以上版本](#32-安装-ndk-20-以上版本)
+    - [3.3 导入项目](#33-导入项目)
+  - [4 更多支持](#4-更多支持)
+
+# Android Demo
+
+## 1. 简介
+此为PaddleOCR的Android Demo，目前支持文本检测，文本方向分类器和文本识别模型的使用。使用 [PaddleLite v2.10](https://github.com/PaddlePaddle/Paddle-Lite/tree/release/v2.10) 进行开发。
+
+
+## 2. 近期更新
+* 2022.02.27 
+    * 预测库更新到PaddleLite v2.10
+    * 支持6种运行模式：
+      * 检测+分类+识别
+      * 检测+识别
+      * 分类+识别
+      * 检测
+      * 识别
+      * 分类
+
+## 3. 快速使用
+
+### 3.1 安装最新版本的Android Studio
 可以从 https://developer.android.com/studio 下载。本Demo使用是4.0版本Android Studio编写。
 
-### 2. 按照NDK 20 以上版本
+### 3.2 安装 NDK 20 以上版本
 Demo测试的时候使用的是NDK 20b版本，20版本以上均可以支持编译成功。
 
 如果您是初学者，可以用以下方式安装和测试NDK编译环境。
 点击 File -> New ->New Project，  新建  "Native C++" project
 
-### 3. 导入项目
+### 3.3 导入项目
+
 点击 File->New->Import Project...， 然后跟着Android Studio的引导导入
 
+## 4 更多支持
 
-# 获得更多支持
-前往[端计算模型生成平台EasyEdge](https://ai.baidu.com/easyedge/app/open_source_demo?referrerUrl=paddlelite)，获得更多开发支持：
+前往[Paddle-Lite](https://github.com/PaddlePaddle/Paddle-Lite)，获得更多开发支持
 
-- Demo APP：可使用手机扫码安装，方便手机端快速体验文字识别
-- SDK：模型被封装为适配不同芯片硬件和操作系统SDK，包括完善的接口，方便进行二次开发

deploy/android_demo/app/build.gradle

@@ -8,8 +8,8 @@ android {
         applicationId "com.baidu.paddle.lite.demo.ocr"
         minSdkVersion 23
         targetSdkVersion 29
-        versionCode 1
-        versionName "1.0"
+        versionCode 2
+        versionName "2.0"
         testInstrumentationRunner "android.support.test.runner.AndroidJUnitRunner"
         externalNativeBuild {
             cmake {
@@ -17,11 +17,6 @@ android {
                 arguments '-DANDROID_PLATFORM=android-23', '-DANDROID_STL=c++_shared' ,"-DANDROID_ARM_NEON=TRUE"
             }
         }
-        ndk {
-            // abiFilters "arm64-v8a", "armeabi-v7a"
-            abiFilters   "arm64-v8a", "armeabi-v7a"
-            ldLibs "jnigraphics"
-        }
     }
     buildTypes {
         release {
@@ -48,7 +43,7 @@ dependencies {
 
 def archives = [
         [
-                'src' : 'https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/paddle_lite_libs_v2_9_0.tar.gz',
+                'src' : 'https://paddleocr.bj.bcebos.com/libs/paddle_lite_libs_v2_10.tar.gz',
                 'dest': 'PaddleLite'
         ],
         [
@@ -56,7 +51,7 @@ def archives = [
                 'dest': 'OpenCV'
         ],
         [
-                'src' : 'https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ocr_v2_for_cpu.tar.gz',
+                'src' : 'https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2.tar.gz',
                 'dest' : 'src/main/assets/models'
         ],
         [

deploy/android_demo/app/src/main/AndroidManifest.xml

@@ -14,7 +14,6 @@
             android:roundIcon="@mipmap/ic_launcher_round"
             android:supportsRtl="true"
             android:theme="@style/AppTheme">
-        <!-- to test MiniActivity, change this to com.baidu.paddle.lite.demo.ocr.MiniActivity -->
         <activity android:name="com.baidu.paddle.lite.demo.ocr.MainActivity">
             <intent-filter>
                 <action android:name="android.intent.action.MAIN"/>

deploy/android_demo/app/src/main/cpp/native.cpp

@@ -13,7 +13,7 @@ static paddle::lite_api::PowerMode str_to_cpu_mode(const std::string &cpu_mode);
 extern "C" JNIEXPORT jlong JNICALL
 Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_init(
     JNIEnv *env, jobject thiz, jstring j_det_model_path,
-    jstring j_rec_model_path, jstring j_cls_model_path, jint j_thread_num,
+    jstring j_rec_model_path, jstring j_cls_model_path, jint j_use_opencl, jint j_thread_num,
     jstring j_cpu_mode) {
   std::string det_model_path = jstring_to_cpp_string(env, j_det_model_path);
   std::string rec_model_path = jstring_to_cpp_string(env, j_rec_model_path);
@@ -21,6 +21,7 @@ Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_init(
   int thread_num = j_thread_num;
   std::string cpu_mode = jstring_to_cpp_string(env, j_cpu_mode);
   ppredictor::OCR_Config conf;
+  conf.use_opencl = j_use_opencl;
   conf.thread_num = thread_num;
   conf.mode = str_to_cpu_mode(cpu_mode);
   ppredictor::OCR_PPredictor *orc_predictor =
@@ -57,32 +58,31 @@ str_to_cpu_mode(const std::string &cpu_mode) {
 
 extern "C" JNIEXPORT jfloatArray JNICALL
 Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_forward(
-    JNIEnv *env, jobject thiz, jlong java_pointer, jfloatArray buf,
-    jfloatArray ddims, jobject original_image) {
+    JNIEnv *env, jobject thiz, jlong java_pointer, jobject original_image,jint j_max_size_len, jint j_run_det, jint j_run_cls, jint j_run_rec) {
   LOGI("begin to run native forward");
   if (java_pointer == 0) {
     LOGE("JAVA pointer is NULL");
     return cpp_array_to_jfloatarray(env, nullptr, 0);
   }
+
   cv::Mat origin = bitmap_to_cv_mat(env, original_image);
   if (origin.size == 0) {
     LOGE("origin bitmap cannot convert to CV Mat");
     return cpp_array_to_jfloatarray(env, nullptr, 0);
   }
+
+  int max_size_len = j_max_size_len;
+  int run_det = j_run_det;
+  int run_cls = j_run_cls;
+  int run_rec = j_run_rec;
+
   ppredictor::OCR_PPredictor *ppredictor =
       (ppredictor::OCR_PPredictor *)java_pointer;
-  std::vector<float> dims_float_arr = jfloatarray_to_float_vector(env, ddims);
   std::vector<int64_t> dims_arr;
-  dims_arr.resize(dims_float_arr.size());
-  std::copy(dims_float_arr.cbegin(), dims_float_arr.cend(), dims_arr.begin());
-
-  // 这里值有点大，就不调用jfloatarray_to_float_vector了
-  int64_t buf_len = (int64_t)env->GetArrayLength(buf);
-  jfloat *buf_data = env->GetFloatArrayElements(buf, JNI_FALSE);
-  float *data = (jfloat *)buf_data;
   std::vector<ppredictor::OCRPredictResult> results =
-      ppredictor->infer_ocr(dims_arr, data, buf_len, NET_OCR, origin);
+      ppredictor->infer_ocr(origin, max_size_len, run_det, run_cls, run_rec);
   LOGI("infer_ocr finished with boxes %ld", results.size());
+
   // 这里将std::vector<ppredictor::OCRPredictResult> 序列化成
   // float数组，传输到java层再反序列化
   std::vector<float> float_arr;
@@ -90,13 +90,18 @@ Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_forward(
     float_arr.push_back(r.points.size());
     float_arr.push_back(r.word_index.size());
     float_arr.push_back(r.score);
+    // add det point
     for (const std::vector<int> &point : r.points) {
       float_arr.push_back(point.at(0));
       float_arr.push_back(point.at(1));
     }
+    // add rec word idx
     for (int index : r.word_index) {
       float_arr.push_back(index);
     }
+    // add cls result
+    float_arr.push_back(r.cls_label);
+    float_arr.push_back(r.cls_score);
   }
   return cpp_array_to_jfloatarray(env, float_arr.data(), float_arr.size());
 }

deploy/android_demo/app/src/main/cpp/ocr_ppredictor.cpp

@@ -17,15 +17,15 @@ int OCR_PPredictor::init(const std::string &det_model_content,
                          const std::string &rec_model_content,
                          const std::string &cls_model_content) {
   _det_predictor = std::unique_ptr<PPredictor>(
-      new PPredictor{_config.thread_num, NET_OCR, _config.mode});
+      new PPredictor{_config.use_opencl,_config.thread_num, NET_OCR, _config.mode});
   _det_predictor->init_nb(det_model_content);
 
   _rec_predictor = std::unique_ptr<PPredictor>(
-      new PPredictor{_config.thread_num, NET_OCR_INTERNAL, _config.mode});
+      new PPredictor{_config.use_opencl,_config.thread_num, NET_OCR_INTERNAL, _config.mode});
   _rec_predictor->init_nb(rec_model_content);
 
   _cls_predictor = std::unique_ptr<PPredictor>(
-      new PPredictor{_config.thread_num, NET_OCR_INTERNAL, _config.mode});
+      new PPredictor{_config.use_opencl,_config.thread_num, NET_OCR_INTERNAL, _config.mode});
   _cls_predictor->init_nb(cls_model_content);
   return RETURN_OK;
 }
@@ -34,15 +34,16 @@ int OCR_PPredictor::init_from_file(const std::string &det_model_path,
                                    const std::string &rec_model_path,
                                    const std::string &cls_model_path) {
   _det_predictor = std::unique_ptr<PPredictor>(
-      new PPredictor{_config.thread_num, NET_OCR, _config.mode});
+      new PPredictor{_config.use_opencl, _config.thread_num, NET_OCR, _config.mode});
   _det_predictor->init_from_file(det_model_path);
 
+
   _rec_predictor = std::unique_ptr<PPredictor>(
-      new PPredictor{_config.thread_num, NET_OCR_INTERNAL, _config.mode});
+      new PPredictor{_config.use_opencl,_config.thread_num, NET_OCR_INTERNAL, _config.mode});
   _rec_predictor->init_from_file(rec_model_path);
 
   _cls_predictor = std::unique_ptr<PPredictor>(
-      new PPredictor{_config.thread_num, NET_OCR_INTERNAL, _config.mode});
+      new PPredictor{_config.use_opencl,_config.thread_num, NET_OCR_INTERNAL, _config.mode});
   _cls_predictor->init_from_file(cls_model_path);
   return RETURN_OK;
 }
@@ -77,90 +78,173 @@ visual_img(const std::vector<std::vector<std::vector<int>>> &filter_boxes,
 }
 
 std::vector<OCRPredictResult>
-OCR_PPredictor::infer_ocr(const std::vector<int64_t> &dims,
-                          const float *input_data, int input_len, int net_flag,
-                          cv::Mat &origin) {
+OCR_PPredictor::infer_ocr(cv::Mat &origin,int max_size_len, int run_det, int run_cls, int run_rec) {
+  LOGI("ocr cpp start *****************");
+  LOGI("ocr cpp det: %d, cls: %d, rec: %d", run_det, run_cls, run_rec);
+  std::vector<OCRPredictResult> ocr_results;
+  if(run_det){
+    infer_det(origin, max_size_len, ocr_results);
+  }
+  if(run_rec){
+    if(ocr_results.size()==0){
+      OCRPredictResult res;
+      ocr_results.emplace_back(std::move(res));
+    }
+    for(int i = 0; i < ocr_results.size();i++) {
+      infer_rec(origin, run_cls, ocr_results[i]);
+    }
+  }else if(run_cls){
+    ClsPredictResult cls_res = infer_cls(origin);
+    OCRPredictResult res;
+    res.cls_score = cls_res.cls_score;
+    res.cls_label = cls_res.cls_label;
+    ocr_results.push_back(res);
+  }
+
+  LOGI("ocr cpp end *****************");
+  return ocr_results;
+}
+
+cv::Mat DetResizeImg(const cv::Mat img, int max_size_len,
+                     std::vector<float> &ratio_hw) {
+  int w = img.cols;
+  int h = img.rows;
+
+  float ratio = 1.f;
+  int max_wh = w >= h ? w : h;
+  if (max_wh > max_size_len) {
+    if (h > w) {
+      ratio = static_cast<float>(max_size_len) / static_cast<float>(h);
+    } else {
+      ratio = static_cast<float>(max_size_len) / static_cast<float>(w);
+    }
+  }
+
+  int resize_h = static_cast<int>(float(h) * ratio);
+  int resize_w = static_cast<int>(float(w) * ratio);
+  if (resize_h % 32 == 0)
+    resize_h = resize_h;
+  else if (resize_h / 32 < 1 + 1e-5)
+    resize_h = 32;
+  else
+    resize_h = (resize_h / 32 - 1) * 32;
+
+  if (resize_w % 32 == 0)
+    resize_w = resize_w;
+  else if (resize_w / 32 < 1 + 1e-5)
+    resize_w = 32;
+  else
+    resize_w = (resize_w / 32 - 1) * 32;
+
+  cv::Mat resize_img;
+  cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
+
+  ratio_hw.push_back(static_cast<float>(resize_h) / static_cast<float>(h));
+  ratio_hw.push_back(static_cast<float>(resize_w) / static_cast<float>(w));
+  return resize_img;
+}
+
+void OCR_PPredictor::infer_det(cv::Mat &origin, int max_size_len, std::vector<OCRPredictResult> &ocr_results) {
+  std::vector<float> mean = {0.485f, 0.456f, 0.406f};
+  std::vector<float> scale = {1 / 0.229f, 1 / 0.224f, 1 / 0.225f};
+
   PredictorInput input = _det_predictor->get_first_input();
-  input.set_dims(dims);
-  input.set_data(input_data, input_len);
+
+  std::vector<float> ratio_hw;
+  cv::Mat input_image = DetResizeImg(origin, max_size_len, ratio_hw);
+  input_image.convertTo(input_image, CV_32FC3, 1 / 255.0f);
+  const float *dimg = reinterpret_cast<const float *>(input_image.data);
+  int input_size = input_image.rows * input_image.cols;
+
+  input.set_dims({1, 3, input_image.rows, input_image.cols});
+
+  neon_mean_scale(dimg, input.get_mutable_float_data(), input_size, mean,
+                  scale);
+  LOGI("ocr cpp det shape %d,%d", input_image.rows,input_image.cols);
   std::vector<PredictorOutput> results = _det_predictor->infer();
   PredictorOutput &res = results.at(0);
   std::vector<std::vector<std::vector<int>>> filtered_box = calc_filtered_boxes(
-      res.get_float_data(), res.get_size(), (int)dims[2], (int)dims[3], origin);
-  LOGI("Filter_box size %ld", filtered_box.size());
-  return infer_rec(filtered_box, origin);
+          res.get_float_data(), res.get_size(), input_image.rows, input_image.cols, origin);
+  LOGI("ocr cpp det Filter_box size %ld", filtered_box.size());
+
+  for(int i = 0;i<filtered_box.size();i++){
+    LOGI("ocr cpp box  %d,%d,%d,%d,%d,%d,%d,%d", filtered_box[i][0][0],filtered_box[i][0][1], filtered_box[i][1][0],filtered_box[i][1][1], filtered_box[i][2][0],filtered_box[i][2][1], filtered_box[i][3][0],filtered_box[i][3][1]);
+    OCRPredictResult res;
+    res.points = filtered_box[i];
+    ocr_results.push_back(res);
+  }
 }
 
-std::vector<OCRPredictResult> OCR_PPredictor::infer_rec(
-    const std::vector<std::vector<std::vector<int>>> &boxes,
-    const cv::Mat &origin_img) {
+void OCR_PPredictor::infer_rec(const cv::Mat &origin_img, int run_cls, OCRPredictResult& ocr_result) {
   std::vector<float> mean = {0.5f, 0.5f, 0.5f};
   std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
   std::vector<int64_t> dims = {1, 3, 0, 0};
-  std::vector<OCRPredictResult> ocr_results;
 
   PredictorInput input = _rec_predictor->get_first_input();
-  for (auto bp = boxes.crbegin(); bp != boxes.crend(); ++bp) {
-    const std::vector<std::vector<int>> &box = *bp;
-    cv::Mat crop_img = get_rotate_crop_image(origin_img, box);
-    crop_img = infer_cls(crop_img);
 
-    float wh_ratio = float(crop_img.cols) / float(crop_img.rows);
-    cv::Mat input_image = crnn_resize_img(crop_img, wh_ratio);
-    input_image.convertTo(input_image, CV_32FC3, 1 / 255.0f);
-    const float *dimg = reinterpret_cast<const float *>(input_image.data);
-    int input_size = input_image.rows * input_image.cols;
+  const std::vector<std::vector<int>> &box = ocr_result.points;
+  cv::Mat crop_img;
+  if(box.size()>0){
+    crop_img = get_rotate_crop_image(origin_img, box);
+  }
+  else{
+    crop_img = origin_img;
+  }
 
-    dims[2] = input_image.rows;
-    dims[3] = input_image.cols;
-    input.set_dims(dims);
+  if(run_cls){
+    ClsPredictResult cls_res = infer_cls(crop_img);
+    crop_img = cls_res.img;
+    ocr_result.cls_score = cls_res.cls_score;
+    ocr_result.cls_label = cls_res.cls_label;
+  }
 
-    neon_mean_scale(dimg, input.get_mutable_float_data(), input_size, mean,
-                    scale);
 
-    std::vector<PredictorOutput> results = _rec_predictor->infer();
-    const float *predict_batch = results.at(0).get_float_data();
-    const std::vector<int64_t> predict_shape = results.at(0).get_shape();
+  float wh_ratio = float(crop_img.cols) / float(crop_img.rows);
+  cv::Mat input_image = crnn_resize_img(crop_img, wh_ratio);
+  input_image.convertTo(input_image, CV_32FC3, 1 / 255.0f);
+  const float *dimg = reinterpret_cast<const float *>(input_image.data);
+  int input_size = input_image.rows * input_image.cols;
 
-    OCRPredictResult res;
+  dims[2] = input_image.rows;
+  dims[3] = input_image.cols;
+  input.set_dims(dims);
 
-    // ctc decode
-    int argmax_idx;
-    int last_index = 0;
-    float score = 0.f;
-    int count = 0;
-    float max_value = 0.0f;
-
-    for (int n = 0; n < predict_shape[1]; n++) {
-      argmax_idx = int(argmax(&predict_batch[n * predict_shape[2]],
-                              &predict_batch[(n + 1) * predict_shape[2]]));
-      max_value =
-          float(*std::max_element(&predict_batch[n * predict_shape[2]],
-                                  &predict_batch[(n + 1) * predict_shape[2]]));
-      if (argmax_idx > 0 && (!(n > 0 && argmax_idx == last_index))) {
-        score += max_value;
-        count += 1;
-        res.word_index.push_back(argmax_idx);
-      }
-      last_index = argmax_idx;
-    }
-    score /= count;
-    if (res.word_index.empty()) {
-      continue;
+  neon_mean_scale(dimg, input.get_mutable_float_data(), input_size, mean,
+                  scale);
+
+  std::vector<PredictorOutput> results = _rec_predictor->infer();
+  const float *predict_batch = results.at(0).get_float_data();
+  const std::vector<int64_t> predict_shape = results.at(0).get_shape();
+
+  // ctc decode
+  int argmax_idx;
+  int last_index = 0;
+  float score = 0.f;
+  int count = 0;
+  float max_value = 0.0f;
+
+  for (int n = 0; n < predict_shape[1]; n++) {
+    argmax_idx = int(argmax(&predict_batch[n * predict_shape[2]],
+                            &predict_batch[(n + 1) * predict_shape[2]]));
+    max_value =
+        float(*std::max_element(&predict_batch[n * predict_shape[2]],
+                                &predict_batch[(n + 1) * predict_shape[2]]));
+    if (argmax_idx > 0 && (!(n > 0 && argmax_idx == last_index))) {
+      score += max_value;
+      count += 1;
+      ocr_result.word_index.push_back(argmax_idx);
     }
-    res.score = score;
-    res.points = box;
-    ocr_results.emplace_back(std::move(res));
+    last_index = argmax_idx;
   }
-  LOGI("ocr_results finished %lu", ocr_results.size());
-  return ocr_results;
+  score /= count;
+  ocr_result.score = score;
+  LOGI("ocr cpp rec word size %ld", count);
 }
 
-cv::Mat OCR_PPredictor::infer_cls(const cv::Mat &img, float thresh) {
+ClsPredictResult OCR_PPredictor::infer_cls(const cv::Mat &img, float thresh) {
   std::vector<float> mean = {0.5f, 0.5f, 0.5f};
   std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
   std::vector<int64_t> dims = {1, 3, 0, 0};
-  std::vector<OCRPredictResult> ocr_results;
 
   PredictorInput input = _cls_predictor->get_first_input();
 
@@ -182,7 +266,7 @@ cv::Mat OCR_PPredictor::infer_cls(const cv::Mat &img, float thresh) {
   float score = 0;
   int label = 0;
   for (int64_t i = 0; i < results.at(0).get_size(); i++) {
-    LOGI("output scores [%f]", scores[i]);
+    LOGI("ocr cpp cls output scores [%f]", scores[i]);
     if (scores[i] > score) {
       score = scores[i];
       label = i;
@@ -193,7 +277,12 @@ cv::Mat OCR_PPredictor::infer_cls(const cv::Mat &img, float thresh) {
   if (label % 2 == 1 && score > thresh) {
     cv::rotate(srcimg, srcimg, 1);
   }
-  return srcimg;
+  ClsPredictResult res;
+  res.cls_label = label;
+  res.cls_score = score;
+  res.img = srcimg;
+  LOGI("ocr cpp cls word cls %ld, %f", label, score);
+  return res;
 }
 
 std::vector<std::vector<std::vector<int>>>

deploy/android_demo/app/src/main/cpp/ocr_ppredictor.h

@@ -15,7 +15,8 @@ namespace ppredictor {
  * Config
  */
 struct OCR_Config {
-  int thread_num = 4; // Thread num
+    int use_opencl = 0;
+    int thread_num = 4; // Thread num
   paddle::lite_api::PowerMode mode =
       paddle::lite_api::LITE_POWER_HIGH; // PaddleLite Mode
 };
@@ -27,8 +28,15 @@ struct OCRPredictResult {
   std::vector<int> word_index;
   std::vector<std::vector<int>> points;
   float score;
+  float cls_score;
+  int cls_label=-1;
 };
 
+struct ClsPredictResult {
+    float cls_score;
+    int cls_label=-1;
+    cv::Mat img;
+};
 /**
  * OCR there are 2 models
  * 1. First model（det），select polygones to show where are the texts
@@ -62,8 +70,7 @@ public:
    * @return
    */
   virtual std::vector<OCRPredictResult>
-  infer_ocr(const std::vector<int64_t> &dims, const float *input_data,
-            int input_len, int net_flag, cv::Mat &origin);
+  infer_ocr(cv::Mat &origin, int max_size_len, int run_det, int run_cls, int run_rec);
 
   virtual NET_TYPE get_net_flag() const;
 
@@ -80,25 +87,26 @@ private:
   calc_filtered_boxes(const float *pred, int pred_size, int output_height,
                       int output_width, const cv::Mat &origin);
 
+  void
+  infer_det(cv::Mat &origin, int max_side_len, std::vector<OCRPredictResult>& ocr_results);
   /**
-   * infer for second model
+   * infer for rec model
    *
    * @param boxes
    * @param origin
    * @return
    */
-  std::vector<OCRPredictResult>
-  infer_rec(const std::vector<std::vector<std::vector<int>>> &boxes,
-            const cv::Mat &origin);
+  void
+  infer_rec(const cv::Mat &origin, int run_cls, OCRPredictResult& ocr_result);
 
-  /**
+    /**
   * infer for cls model
   *
   * @param boxes
   * @param origin
   * @return
   */
-  cv::Mat infer_cls(const cv::Mat &origin, float thresh = 0.9);
+  ClsPredictResult infer_cls(const cv::Mat &origin, float thresh = 0.9);
 
   /**
    * Postprocess or sencod model to extract text

deploy/android_demo/app/src/main/cpp/ppredictor.cpp

@@ -2,9 +2,9 @@
 #include "common.h"
 
 namespace ppredictor {
-PPredictor::PPredictor(int thread_num, int net_flag,
+PPredictor::PPredictor(int use_opencl, int thread_num, int net_flag,
                        paddle::lite_api::PowerMode mode)
-    : _thread_num(thread_num), _net_flag(net_flag), _mode(mode) {}
+    : _use_opencl(use_opencl), _thread_num(thread_num), _net_flag(net_flag), _mode(mode) {}
 
 int PPredictor::init_nb(const std::string &model_content) {
   paddle::lite_api::MobileConfig config;
@@ -19,10 +19,40 @@ int PPredictor::init_from_file(const std::string &model_content) {
 }
 
 template <typename ConfigT> int PPredictor::_init(ConfigT &config) {
+  bool is_opencl_backend_valid = paddle::lite_api::IsOpenCLBackendValid(/*check_fp16_valid = false*/);
+  if (is_opencl_backend_valid) {
+    if (_use_opencl != 0) {
+      // Make sure you have write permission of the binary path.
+      // We strongly recommend each model has a unique binary name.
+      const std::string bin_path = "/data/local/tmp/";
+      const std::string bin_name = "lite_opencl_kernel.bin";
+      config.set_opencl_binary_path_name(bin_path, bin_name);
+
+      // opencl tune option
+      // CL_TUNE_NONE: 0
+      // CL_TUNE_RAPID: 1
+      // CL_TUNE_NORMAL: 2
+      // CL_TUNE_EXHAUSTIVE: 3
+      const std::string tuned_path = "/data/local/tmp/";
+      const std::string tuned_name = "lite_opencl_tuned.bin";
+      config.set_opencl_tune(paddle::lite_api::CL_TUNE_NORMAL, tuned_path, tuned_name);
+
+      // opencl precision option
+      // CL_PRECISION_AUTO: 0, first fp16 if valid, default
+      // CL_PRECISION_FP32: 1, force fp32
+      // CL_PRECISION_FP16: 2, force fp16
+      config.set_opencl_precision(paddle::lite_api::CL_PRECISION_FP32);
+      LOGI("ocr cpp device: running on gpu.");
+    }
+  } else {
+    LOGI("ocr cpp device: running on cpu.");
+    // you can give backup cpu nb model instead
+    // config.set_model_from_file(cpu_nb_model_dir);
+  }
   config.set_threads(_thread_num);
   config.set_power_mode(_mode);
   _predictor = paddle::lite_api::CreatePaddlePredictor(config);
-  LOGI("paddle instance created");
+  LOGI("ocr cpp paddle instance created");
   return RETURN_OK;
 }
 
@@ -43,18 +73,18 @@ std::vector<PredictorInput> PPredictor::get_inputs(int num) {
 PredictorInput PPredictor::get_first_input() { return get_input(0); }
 
 std::vector<PredictorOutput> PPredictor::infer() {
-  LOGI("infer Run start %d", _net_flag);
+  LOGI("ocr cpp infer Run start %d", _net_flag);
   std::vector<PredictorOutput> results;
   if (!_is_input_get) {
     return results;
   }
   _predictor->Run();
-  LOGI("infer Run end");
+  LOGI("ocr cpp infer Run end");
 
   for (int i = 0; i < _predictor->GetOutputNames().size(); i++) {
     std::unique_ptr<const paddle::lite_api::Tensor> output_tensor =
         _predictor->GetOutput(i);
-    LOGI("output tensor[%d] size %ld", i, product(output_tensor->shape()));
+    LOGI("ocr cpp output tensor[%d] size %ld", i, product(output_tensor->shape()));
     PredictorOutput result{std::move(output_tensor), i, _net_flag};
     results.emplace_back(std::move(result));
   }

deploy/android_demo/app/src/main/cpp/ppredictor.h

@@ -22,7 +22,7 @@ public:
 class PPredictor : public PPredictor_Interface {
 public:
   PPredictor(
-      int thread_num, int net_flag = 0,
+          int use_opencl, int thread_num, int net_flag = 0,
       paddle::lite_api::PowerMode mode = paddle::lite_api::LITE_POWER_HIGH);
 
   virtual ~PPredictor() {}
@@ -54,6 +54,7 @@ protected:
   template <typename ConfigT> int _init(ConfigT &config);
 
 private:
+    int _use_opencl;
   int _thread_num;
   paddle::lite_api::PowerMode _mode;
   std::shared_ptr<paddle::lite_api::PaddlePredictor> _predictor;

deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/MainActivity.java

@@ -13,6 +13,7 @@ import android.graphics.BitmapFactory;
 import android.graphics.drawable.BitmapDrawable;
 import android.media.ExifInterface;
 import android.content.res.AssetManager;
+import android.media.FaceDetector;
 import android.net.Uri;
 import android.os.Bundle;
 import android.os.Environment;
@@ -27,7 +28,9 @@ import android.view.Menu;
 import android.view.MenuInflater;
 import android.view.MenuItem;
 import android.view.View;
+import android.widget.CheckBox;
 import android.widget.ImageView;
+import android.widget.Spinner;
 import android.widget.TextView;
 import android.widget.Toast;
 
@@ -68,23 +71,24 @@ public class MainActivity extends AppCompatActivity {
     protected ImageView ivInputImage;
     protected TextView tvOutputResult;
     protected TextView tvInferenceTime;
+    protected CheckBox cbOpencl;
+    protected Spinner spRunMode;
 
-    // Model settings of object detection
+    // Model settings of ocr
     protected String modelPath = "";
     protected String labelPath = "";
     protected String imagePath = "";
     protected int cpuThreadNum = 1;
     protected String cpuPowerMode = "";
-    protected String inputColorFormat = "";
-    protected long[] inputShape = new long[]{};
-    protected float[] inputMean = new float[]{};
-    protected float[] inputStd = new float[]{};
+    protected int detLongSize = 960;
     protected float scoreThreshold = 0.1f;
     private String currentPhotoPath;
-    private AssetManager assetManager =null;
+    private AssetManager assetManager = null;
 
     protected Predictor predictor = new Predictor();
 
+    private Bitmap cur_predict_image = null;
+
     @Override
     protected void onCreate(Bundle savedInstanceState) {
         super.onCreate(savedInstanceState);
@@ -98,10 +102,12 @@ public class MainActivity extends AppCompatActivity {
 
         // Setup the UI components
         tvInputSetting = findViewById(R.id.tv_input_setting);
+        cbOpencl = findViewById(R.id.cb_opencl);
         tvStatus = findViewById(R.id.tv_model_img_status);
         ivInputImage = findViewById(R.id.iv_input_image);
         tvInferenceTime = findViewById(R.id.tv_inference_time);
         tvOutputResult = findViewById(R.id.tv_output_result);
+        spRunMode = findViewById(R.id.sp_run_mode);
         tvInputSetting.setMovementMethod(ScrollingMovementMethod.getInstance());
         tvOutputResult.setMovementMethod(ScrollingMovementMethod.getInstance());
 
@@ -111,26 +117,26 @@ public class MainActivity extends AppCompatActivity {
             public void handleMessage(Message msg) {
                 switch (msg.what) {
                     case RESPONSE_LOAD_MODEL_SUCCESSED:
-                        if(pbLoadModel!=null && pbLoadModel.isShowing()){
+                        if (pbLoadModel != null && pbLoadModel.isShowing()) {
                             pbLoadModel.dismiss();
                         }
                         onLoadModelSuccessed();
                         break;
                     case RESPONSE_LOAD_MODEL_FAILED:
-                        if(pbLoadModel!=null && pbLoadModel.isShowing()){
+                        if (pbLoadModel != null && pbLoadModel.isShowing()) {
                             pbLoadModel.dismiss();
                         }
                         Toast.makeText(MainActivity.this, "Load model failed!", Toast.LENGTH_SHORT).show();
                         onLoadModelFailed();
                         break;
                     case RESPONSE_RUN_MODEL_SUCCESSED:
-                        if(pbRunModel!=null && pbRunModel.isShowing()){
+                        if (pbRunModel != null && pbRunModel.isShowing()) {
                             pbRunModel.dismiss();
                         }
                         onRunModelSuccessed();
                         break;
                     case RESPONSE_RUN_MODEL_FAILED:
-                        if(pbRunModel!=null && pbRunModel.isShowing()){
+                        if (pbRunModel != null && pbRunModel.isShowing()) {
                             pbRunModel.dismiss();
                         }
                         Toast.makeText(MainActivity.this, "Run model failed!", Toast.LENGTH_SHORT).show();
@@ -175,71 +181,47 @@ public class MainActivity extends AppCompatActivity {
         super.onResume();
         SharedPreferences sharedPreferences = PreferenceManager.getDefaultSharedPreferences(this);
         boolean settingsChanged = false;
+        boolean model_settingsChanged = false;
         String model_path = sharedPreferences.getString(getString(R.string.MODEL_PATH_KEY),
                 getString(R.string.MODEL_PATH_DEFAULT));
         String label_path = sharedPreferences.getString(getString(R.string.LABEL_PATH_KEY),
                 getString(R.string.LABEL_PATH_DEFAULT));
         String image_path = sharedPreferences.getString(getString(R.string.IMAGE_PATH_KEY),
                 getString(R.string.IMAGE_PATH_DEFAULT));
-        settingsChanged |= !model_path.equalsIgnoreCase(modelPath);
+        model_settingsChanged |= !model_path.equalsIgnoreCase(modelPath);
         settingsChanged |= !label_path.equalsIgnoreCase(labelPath);
         settingsChanged |= !image_path.equalsIgnoreCase(imagePath);
         int cpu_thread_num = Integer.parseInt(sharedPreferences.getString(getString(R.string.CPU_THREAD_NUM_KEY),
                 getString(R.string.CPU_THREAD_NUM_DEFAULT)));
-        settingsChanged |= cpu_thread_num != cpuThreadNum;
+        model_settingsChanged |= cpu_thread_num != cpuThreadNum;
         String cpu_power_mode =
                 sharedPreferences.getString(getString(R.string.CPU_POWER_MODE_KEY),
                         getString(R.string.CPU_POWER_MODE_DEFAULT));
-        settingsChanged |= !cpu_power_mode.equalsIgnoreCase(cpuPowerMode);
-        String input_color_format =
-                sharedPreferences.getString(getString(R.string.INPUT_COLOR_FORMAT_KEY),
-                        getString(R.string.INPUT_COLOR_FORMAT_DEFAULT));
-        settingsChanged |= !input_color_format.equalsIgnoreCase(inputColorFormat);
-        long[] input_shape =
-                Utils.parseLongsFromString(sharedPreferences.getString(getString(R.string.INPUT_SHAPE_KEY),
-                        getString(R.string.INPUT_SHAPE_DEFAULT)), ",");
-        float[] input_mean =
-                Utils.parseFloatsFromString(sharedPreferences.getString(getString(R.string.INPUT_MEAN_KEY),
-                        getString(R.string.INPUT_MEAN_DEFAULT)), ",");
-        float[] input_std =
-                Utils.parseFloatsFromString(sharedPreferences.getString(getString(R.string.INPUT_STD_KEY)
-                        , getString(R.string.INPUT_STD_DEFAULT)), ",");
-        settingsChanged |= input_shape.length != inputShape.length;
-        settingsChanged |= input_mean.length != inputMean.length;
-        settingsChanged |= input_std.length != inputStd.length;
-        if (!settingsChanged) {
-            for (int i = 0; i < input_shape.length; i++) {
-                settingsChanged |= input_shape[i] != inputShape[i];
-            }
-            for (int i = 0; i < input_mean.length; i++) {
-                settingsChanged |= input_mean[i] != inputMean[i];
-            }
-            for (int i = 0; i < input_std.length; i++) {
-                settingsChanged |= input_std[i] != inputStd[i];
-            }
-        }
+        model_settingsChanged |= !cpu_power_mode.equalsIgnoreCase(cpuPowerMode);
+
+        int det_long_size = Integer.parseInt(sharedPreferences.getString(getString(R.string.DET_LONG_SIZE_KEY),
+                getString(R.string.DET_LONG_SIZE_DEFAULT)));
+        settingsChanged |= det_long_size != detLongSize;
         float score_threshold =
                 Float.parseFloat(sharedPreferences.getString(getString(R.string.SCORE_THRESHOLD_KEY),
                         getString(R.string.SCORE_THRESHOLD_DEFAULT)));
         settingsChanged |= scoreThreshold != score_threshold;
         if (settingsChanged) {
-            modelPath = model_path;
             labelPath = label_path;
             imagePath = image_path;
+            detLongSize = det_long_size;
+            scoreThreshold = score_threshold;
+            set_img();
+        }
+        if (model_settingsChanged) {
+            modelPath = model_path;
             cpuThreadNum = cpu_thread_num;
             cpuPowerMode = cpu_power_mode;
-            inputColorFormat = input_color_format;
-            inputShape = input_shape;
-            inputMean = input_mean;
-            inputStd = input_std;
-            scoreThreshold = score_threshold;
             // Update UI
-            tvInputSetting.setText("Model: " + modelPath.substring(modelPath.lastIndexOf("/") + 1) + "\n" + "CPU" +
-                    " Thread Num: " + Integer.toString(cpuThreadNum) + "\n" + "CPU Power Mode: " + cpuPowerMode);
+            tvInputSetting.setText("Model: " + modelPath.substring(modelPath.lastIndexOf("/") + 1) + "\nOPENCL: " + cbOpencl.isChecked() + "\nCPU Thread Num: " + cpuThreadNum + "\nCPU Power Mode: " + cpuPowerMode);
             tvInputSetting.scrollTo(0, 0);
             // Reload model if configure has been changed
-//            loadModel();
-            set_img();
+            loadModel();
         }
     }
 
@@ -254,20 +236,28 @@ public class MainActivity extends AppCompatActivity {
     }
 
     public boolean onLoadModel() {
-        return predictor.init(MainActivity.this, modelPath, labelPath, cpuThreadNum,
+        if (predictor.isLoaded()) {
+            predictor.releaseModel();
+        }
+        return predictor.init(MainActivity.this, modelPath, labelPath, cbOpencl.isChecked() ? 1 : 0, cpuThreadNum,
                 cpuPowerMode,
-                inputColorFormat,
-                inputShape, inputMean,
-                inputStd, scoreThreshold);
+                detLongSize, scoreThreshold);
     }
 
     public boolean onRunModel() {
-        return predictor.isLoaded() && predictor.runModel();
+        String run_mode = spRunMode.getSelectedItem().toString();
+        int run_det = run_mode.contains("检测") ? 1 : 0;
+        int run_cls = run_mode.contains("分类") ? 1 : 0;
+        int run_rec = run_mode.contains("识别") ? 1 : 0;
+        return predictor.isLoaded() && predictor.runModel(run_det, run_cls, run_rec);
     }
 
     public void onLoadModelSuccessed() {
         // Load test image from path and run model
+        tvInputSetting.setText("Model: " + modelPath.substring(modelPath.lastIndexOf("/") + 1) + "\nOPENCL: " + cbOpencl.isChecked() + "\nCPU Thread Num: " + cpuThreadNum + "\nCPU Power Mode: " + cpuPowerMode);
+        tvInputSetting.scrollTo(0, 0);
         tvStatus.setText("STATUS: load model successed");
+
     }
 
     public void onLoadModelFailed() {
@@ -290,20 +280,13 @@ public class MainActivity extends AppCompatActivity {
         tvStatus.setText("STATUS: run model failed");
     }
 
-    public void onImageChanged(Bitmap image) {
-        // Rerun model if users pick test image from gallery or camera
-        if (image != null && predictor.isLoaded()) {
-            predictor.setInputImage(image);
-            runModel();
-        }
-    }
-
     public void set_img() {
         // Load test image from path and run model
         try {
-            assetManager= getAssets();
-            InputStream in=assetManager.open(imagePath);
-            Bitmap bmp=BitmapFactory.decodeStream(in);
+            assetManager = getAssets();
+            InputStream in = assetManager.open(imagePath);
+            Bitmap bmp = BitmapFactory.decodeStream(in);
+            cur_predict_image = bmp;
             ivInputImage.setImageBitmap(bmp);
         } catch (IOException e) {
             Toast.makeText(MainActivity.this, "Load image failed!", Toast.LENGTH_SHORT).show();
@@ -430,7 +413,7 @@ public class MainActivity extends AppCompatActivity {
                         Cursor cursor = managedQuery(uri, proj, null, null, null);
                         cursor.moveToFirst();
                         if (image != null) {
-//                            onImageChanged(image);
+                            cur_predict_image = image;
                             ivInputImage.setImageBitmap(image);
                         }
                     } catch (IOException e) {
@@ -451,7 +434,7 @@ public class MainActivity extends AppCompatActivity {
                         Bitmap image = BitmapFactory.decodeFile(currentPhotoPath);
                         image = Utils.rotateBitmap(image, orientation);
                         if (image != null) {
-//                            onImageChanged(image);
+                            cur_predict_image = image;
                             ivInputImage.setImageBitmap(image);
                         }
                     } else {
@@ -464,28 +447,28 @@ public class MainActivity extends AppCompatActivity {
         }
     }
 
-    public void btn_load_model_click(View view) {
-        if (predictor.isLoaded()){
-            tvStatus.setText("STATUS: model has been loaded");
-        }else{
-            tvStatus.setText("STATUS: load model ......");
-            loadModel();
-        }
+    public void btn_reset_img_click(View view) {
+        ivInputImage.setImageBitmap(cur_predict_image);
+    }
+
+    public void cb_opencl_click(View view) {
+        tvStatus.setText("STATUS: load model ......");
+        loadModel();
     }
 
     public void btn_run_model_click(View view) {
-        Bitmap image =((BitmapDrawable)ivInputImage.getDrawable()).getBitmap();
-        if(image == null) {
+        Bitmap image = ((BitmapDrawable) ivInputImage.getDrawable()).getBitmap();
+        if (image == null) {
             tvStatus.setText("STATUS: image is not exists");
-        }
-        else if (!predictor.isLoaded()){
+        } else if (!predictor.isLoaded()) {
             tvStatus.setText("STATUS: model is not loaded");
-        }else{
+        } else {
             tvStatus.setText("STATUS: run model ...... ");
             predictor.setInputImage(image);
             runModel();
         }
     }
+
     public void btn_choice_img_click(View view) {
         if (requestAllPermissions()) {
             openGallery();
@@ -506,4 +489,32 @@ public class MainActivity extends AppCompatActivity {
         worker.quit();
         super.onDestroy();
     }
+
+    public int get_run_mode() {
+        String run_mode = spRunMode.getSelectedItem().toString();
+        int mode;
+        switch (run_mode) {
+            case "检测+分类+识别":
+                mode = 1;
+                break;
+            case "检测+识别":
+                mode = 2;
+                break;
+            case "识别+分类":
+                mode = 3;
+                break;
+            case "检测":
+                mode = 4;
+                break;
+            case "识别":
+                mode = 5;
+                break;
+            case "分类":
+                mode = 6;
+                break;
+            default:
+                mode = 1;
+        }
+        return mode;
+    }
 }

deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/MiniActivity.java

-package com.baidu.paddle.lite.demo.ocr;
-
-import android.graphics.Bitmap;
-import android.graphics.BitmapFactory;
-import android.os.Build;
-import android.os.Bundle;
-import android.os.Handler;
-import android.os.HandlerThread;
-import android.os.Message;
-import android.util.Log;
-import android.view.View;
-import android.widget.Button;
-import android.widget.ImageView;
-import android.widget.TextView;
-import android.widget.Toast;
-
-import androidx.appcompat.app.AppCompatActivity;
-
-import java.io.IOException;
-import java.io.InputStream;
-
-public class MiniActivity extends AppCompatActivity {
-
-
-    public static final int REQUEST_LOAD_MODEL = 0;
-    public static final int REQUEST_RUN_MODEL = 1;
-    public static final int REQUEST_UNLOAD_MODEL = 2;
-    public static final int RESPONSE_LOAD_MODEL_SUCCESSED = 0;
-    public static final int RESPONSE_LOAD_MODEL_FAILED = 1;
-    public static final int RESPONSE_RUN_MODEL_SUCCESSED = 2;
-    public static final int RESPONSE_RUN_MODEL_FAILED = 3;
-
-    private static final String TAG = "MiniActivity";
-
-    protected Handler receiver = null; // Receive messages from worker thread
-    protected Handler sender = null; // Send command to worker thread
-    protected HandlerThread worker = null; // Worker thread to load&run model
-    protected volatile Predictor predictor = null;
-
-    private String assetModelDirPath = "models/ocr_v2_for_cpu";
-    private String assetlabelFilePath = "labels/ppocr_keys_v1.txt";
-
-    private Button button;
-    private ImageView imageView; // image result
-    private TextView textView; // text result
-
-    @Override
-    protected void onCreate(Bundle savedInstanceState) {
-        super.onCreate(savedInstanceState);
-        setContentView(R.layout.activity_mini);
-
-        Log.i(TAG, "SHOW in Logcat");
-
-        // Prepare the worker thread for mode loading and inference
-        worker = new HandlerThread("Predictor Worker");
-        worker.start();
-        sender = new Handler(worker.getLooper()) {
-            public void handleMessage(Message msg) {
-                switch (msg.what) {
-                    case REQUEST_LOAD_MODEL:
-                        // Load model and reload test image
-                        if (!onLoadModel()) {
-                            runOnUiThread(new Runnable() {
-                                @Override
-                                public void run() {
-                                    Toast.makeText(MiniActivity.this, "Load model failed!", Toast.LENGTH_SHORT).show();
-                                }
-                            });
-                        }
-                        break;
-                    case REQUEST_RUN_MODEL:
-                        // Run model if model is loaded
-                        final boolean isSuccessed = onRunModel();
-                        runOnUiThread(new Runnable() {
-                            @Override
-                            public void run() {
-                                if (isSuccessed){
-                                    onRunModelSuccessed();
-                                }else{
-                                    Toast.makeText(MiniActivity.this, "Run model failed!", Toast.LENGTH_SHORT).show();
-                                }
-                            }
-                        });
-                        break;
-                }
-            }
-        };
-        sender.sendEmptyMessage(REQUEST_LOAD_MODEL); // corresponding to REQUEST_LOAD_MODEL， to call onLoadModel()
-
-        imageView = findViewById(R.id.imageView);
-        textView = findViewById(R.id.sample_text);
-        button = findViewById(R.id.button);
-        button.setOnClickListener(new View.OnClickListener() {
-            @Override
-            public void onClick(View v) {
-                sender.sendEmptyMessage(REQUEST_RUN_MODEL);
-            }
-        });
-
-
-    }
-
-    @Override
-    protected void onDestroy() {
-        onUnloadModel();
-        if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.JELLY_BEAN_MR2) {
-            worker.quitSafely();
-        } else {
-            worker.quit();
-        }
-        super.onDestroy();
-    }
-
-    /**
-     * call in onCreate, model init
-     *
-     * @return
-     */
-    private boolean onLoadModel() {
-        if (predictor == null) {
-            predictor = new Predictor();
-        }
-        return predictor.init(this, assetModelDirPath, assetlabelFilePath);
-    }
-
-    /**
-     * init engine
-     * call in onCreate
-     *
-     * @return
-     */
-    private boolean onRunModel() {
-        try {
-            String assetImagePath = "images/0.jpg";
-            InputStream imageStream = getAssets().open(assetImagePath);
-            Bitmap image = BitmapFactory.decodeStream(imageStream);
-            // Input is Bitmap
-            predictor.setInputImage(image);
-            return predictor.isLoaded() && predictor.runModel();
-        } catch (IOException e) {
-            e.printStackTrace();
-            return false;
-        }
-    }
-
-    private void onRunModelSuccessed() {
-        Log.i(TAG, "onRunModelSuccessed");
-        textView.setText(predictor.outputResult);
-        imageView.setImageBitmap(predictor.outputImage);
-    }
-
-    private void onUnloadModel() {
-        if (predictor != null) {
-            predictor.releaseModel();
-        }
-    }
-}

deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OCRPredictorNative.java

@@ -29,22 +29,22 @@ public class OCRPredictorNative {
     public OCRPredictorNative(Config config) {
         this.config = config;
         loadLibrary();
-        nativePointer = init(config.detModelFilename, config.recModelFilename,config.clsModelFilename,
+        nativePointer = init(config.detModelFilename, config.recModelFilename, config.clsModelFilename, config.useOpencl,
                 config.cpuThreadNum, config.cpuPower);
         Log.i("OCRPredictorNative", "load success " + nativePointer);
 
     }
 
 
-    public ArrayList<OcrResultModel> runImage(float[] inputData, int width, int height, int channels, Bitmap originalImage) {
-        Log.i("OCRPredictorNative", "begin to run image " + inputData.length + " " + width + " " + height);
-        float[] dims = new float[]{1, channels, height, width};
-        float[] rawResults = forward(nativePointer, inputData, dims, originalImage);
+    public ArrayList<OcrResultModel> runImage(Bitmap originalImage, int max_size_len, int run_det, int run_cls, int run_rec) {
+        Log.i("OCRPredictorNative", "begin to run image ");
+        float[] rawResults = forward(nativePointer, originalImage, max_size_len, run_det, run_cls, run_rec);
         ArrayList<OcrResultModel> results = postprocess(rawResults);
         return results;
     }
 
     public static class Config {
+        public int useOpencl;
         public int cpuThreadNum;
         public String cpuPower;
         public String detModelFilename;
@@ -53,16 +53,16 @@ public class OCRPredictorNative {
 
     }
 
-    public void destory(){
+    public void destory() {
         if (nativePointer > 0) {
             release(nativePointer);
             nativePointer = 0;
         }
     }
 
-    protected native long init(String detModelPath, String recModelPath,String clsModelPath, int threadNum, String cpuMode);
+    protected native long init(String detModelPath, String recModelPath, String clsModelPath, int useOpencl, int threadNum, String cpuMode);
 
-    protected native float[] forward(long pointer, float[] buf, float[] ddims, Bitmap originalImage);
+    protected native float[] forward(long pointer, Bitmap originalImage,int max_size_len, int run_det, int run_cls, int run_rec);
 
     protected native void release(long pointer);
 
@@ -73,9 +73,9 @@ public class OCRPredictorNative {
         while (begin < raw.length) {
             int point_num = Math.round(raw[begin]);
             int word_num = Math.round(raw[begin + 1]);
-            OcrResultModel model = parse(raw, begin + 2, point_num, word_num);
-            begin += 2 + 1 + point_num * 2 + word_num;
-            results.add(model);
+            OcrResultModel res = parse(raw, begin + 2, point_num, word_num);
+            begin += 2 + 1 + point_num * 2 + word_num + 2;
+            results.add(res);
         }
 
         return results;
@@ -83,19 +83,22 @@ public class OCRPredictorNative {
 
     private OcrResultModel parse(float[] raw, int begin, int pointNum, int wordNum) {
         int current = begin;
-        OcrResultModel model = new OcrResultModel();
-        model.setConfidence(raw[current]);
+        OcrResultModel res = new OcrResultModel();
+        res.setConfidence(raw[current]);
         current++;
         for (int i = 0; i < pointNum; i++) {
-            model.addPoints(Math.round(raw[current + i * 2]), Math.round(raw[current + i * 2 + 1]));
+            res.addPoints(Math.round(raw[current + i * 2]), Math.round(raw[current + i * 2 + 1]));
         }
         current += (pointNum * 2);
         for (int i = 0; i < wordNum; i++) {
             int index = Math.round(raw[current + i]);
-            model.addWordIndex(index);
+            res.addWordIndex(index);
         }
+        current += wordNum;
+        res.setClsIdx(raw[current]);
+        res.setClsConfidence(raw[current + 1]);
         Log.i("OCRPredictorNative", "word finished " + wordNum);
-        return model;
+        return res;
     }
 
 

deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OcrResultModel.java

@@ -10,6 +10,9 @@ public class OcrResultModel {
     private List<Integer> wordIndex;
     private String label;
     private float confidence;
+    private float cls_idx;
+    private String cls_label;
+    private float cls_confidence;
 
     public OcrResultModel() {
         super();
@@ -49,4 +52,28 @@ public class OcrResultModel {
     public void setConfidence(float confidence) {
         this.confidence = confidence;
     }
+
+    public float getClsIdx() {
+        return cls_idx;
+    }
+
+    public void setClsIdx(float idx) {
+        this.cls_idx = idx;
+    }
+
+    public String getClsLabel() {
+        return cls_label;
+    }
+
+    public void setClsLabel(String label) {
+        this.cls_label = label;
+    }
+
+    public float getClsConfidence() {
+        return cls_confidence;
+    }
+
+    public void setClsConfidence(float confidence) {
+        this.cls_confidence = confidence;
+    }
 }

deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java

@@ -31,23 +31,19 @@ public class Predictor {
     protected float inferenceTime = 0;
     // Only for object detection
     protected Vector<String> wordLabels = new Vector<String>();
-    protected String inputColorFormat = "BGR";
-    protected long[] inputShape = new long[]{1, 3, 960};
-    protected float[] inputMean = new float[]{0.485f, 0.456f, 0.406f};
-    protected float[] inputStd = new float[]{1.0f / 0.229f, 1.0f / 0.224f, 1.0f / 0.225f};
+    protected int detLongSize = 960;
     protected float scoreThreshold = 0.1f;
     protected Bitmap inputImage = null;
     protected Bitmap outputImage = null;
     protected volatile String outputResult = "";
-    protected float preprocessTime = 0;
     protected float postprocessTime = 0;
 
 
     public Predictor() {
     }
 
-    public boolean init(Context appCtx, String modelPath, String labelPath) {
-        isLoaded = loadModel(appCtx, modelPath, cpuThreadNum, cpuPowerMode);
+    public boolean init(Context appCtx, String modelPath, String labelPath, int useOpencl, int cpuThreadNum, String cpuPowerMode) {
+        isLoaded = loadModel(appCtx, modelPath, useOpencl, cpuThreadNum, cpuPowerMode);
         if (!isLoaded) {
             return false;
         }
@@ -56,49 +52,18 @@ public class Predictor {
     }
 
 
-    public boolean init(Context appCtx, String modelPath, String labelPath, int cpuThreadNum, String cpuPowerMode,
-                        String inputColorFormat,
-                        long[] inputShape, float[] inputMean,
-                        float[] inputStd, float scoreThreshold) {
-        if (inputShape.length != 3) {
-            Log.e(TAG, "Size of input shape should be: 3");
-            return false;
-        }
-        if (inputMean.length != inputShape[1]) {
-            Log.e(TAG, "Size of input mean should be: " + Long.toString(inputShape[1]));
-            return false;
-        }
-        if (inputStd.length != inputShape[1]) {
-            Log.e(TAG, "Size of input std should be: " + Long.toString(inputShape[1]));
-            return false;
-        }
-        if (inputShape[0] != 1) {
-            Log.e(TAG, "Only one batch is supported in the image classification demo, you can use any batch size in " +
-                    "your Apps!");
-            return false;
-        }
-        if (inputShape[1] != 1 && inputShape[1] != 3) {
-            Log.e(TAG, "Only one/three channels are supported in the image classification demo, you can use any " +
-                    "channel size in your Apps!");
-            return false;
-        }
-        if (!inputColorFormat.equalsIgnoreCase("BGR")) {
-            Log.e(TAG, "Only  BGR color format is supported.");
-            return false;
-        }
-        boolean isLoaded = init(appCtx, modelPath, labelPath);
+    public boolean init(Context appCtx, String modelPath, String labelPath, int useOpencl, int cpuThreadNum, String cpuPowerMode,
+                        int detLongSize, float scoreThreshold) {
+        boolean isLoaded = init(appCtx, modelPath, labelPath, useOpencl, cpuThreadNum, cpuPowerMode);
         if (!isLoaded) {
             return false;
         }
-        this.inputColorFormat = inputColorFormat;
-        this.inputShape = inputShape;
-        this.inputMean = inputMean;
-        this.inputStd = inputStd;
+        this.detLongSize = detLongSize;
         this.scoreThreshold = scoreThreshold;
         return true;
     }
 
-    protected boolean loadModel(Context appCtx, String modelPath, int cpuThreadNum, String cpuPowerMode) {
+    protected boolean loadModel(Context appCtx, String modelPath, int useOpencl, int cpuThreadNum, String cpuPowerMode) {
         // Release model if exists
         releaseModel();
 
@@ -118,12 +83,13 @@ public class Predictor {
         }
 
         OCRPredictorNative.Config config = new OCRPredictorNative.Config();
+        config.useOpencl = useOpencl;
         config.cpuThreadNum = cpuThreadNum;
-        config.detModelFilename = realPath + File.separator + "ch_ppocr_mobile_v2.0_det_opt.nb";
-        config.recModelFilename = realPath + File.separator + "ch_ppocr_mobile_v2.0_rec_opt.nb";
-        config.clsModelFilename = realPath + File.separator + "ch_ppocr_mobile_v2.0_cls_opt.nb";
-        Log.e("Predictor", "model path" + config.detModelFilename + " ; " + config.recModelFilename + ";" + config.clsModelFilename);
         config.cpuPower = cpuPowerMode;
+        config.detModelFilename = realPath + File.separator + "det_db.nb";
+        config.recModelFilename = realPath + File.separator + "rec_crnn.nb";
+        config.clsModelFilename = realPath + File.separator + "cls.nb";
+        Log.i("Predictor", "model path" + config.detModelFilename + " ; " + config.recModelFilename + ";" + config.clsModelFilename);
         paddlePredictor = new OCRPredictorNative(config);
 
         this.cpuThreadNum = cpuThreadNum;
@@ -170,82 +136,29 @@ public class Predictor {
     }
 
 
-    public boolean runModel() {
+    public boolean runModel(int run_det, int run_cls, int run_rec) {
         if (inputImage == null || !isLoaded()) {
             return false;
         }
 
-        // Pre-process image, and feed input tensor with pre-processed data
-
-        Bitmap scaleImage = Utils.resizeWithStep(inputImage, Long.valueOf(inputShape[2]).intValue(), 32);
-
-        Date start = new Date();
-        int channels = (int) inputShape[1];
-        int width = scaleImage.getWidth();
-        int height = scaleImage.getHeight();
-        float[] inputData = new float[channels * width * height];
-        if (channels == 3) {
-            int[] channelIdx = null;
-            if (inputColorFormat.equalsIgnoreCase("RGB")) {
-                channelIdx = new int[]{0, 1, 2};
-            } else if (inputColorFormat.equalsIgnoreCase("BGR")) {
-                channelIdx = new int[]{2, 1, 0};
-            } else {
-                Log.i(TAG, "Unknown color format " + inputColorFormat + ", only RGB and BGR color format is " +
-                        "supported!");
-                return false;
-            }
-
-            int[] channelStride = new int[]{width * height, width * height * 2};
-            int[] pixels=new int[width*height];
-            scaleImage.getPixels(pixels,0,scaleImage.getWidth(),0,0,scaleImage.getWidth(),scaleImage.getHeight());
-            for (int i = 0; i < pixels.length; i++) {
-                int color = pixels[i];
-                float[] rgb = new float[]{(float) red(color) / 255.0f, (float) green(color) / 255.0f,
-                        (float) blue(color) / 255.0f};
-                inputData[i] = (rgb[channelIdx[0]] - inputMean[0]) / inputStd[0];
-                inputData[i + channelStride[0]] = (rgb[channelIdx[1]] - inputMean[1]) / inputStd[1];
-                inputData[i+ channelStride[1]] = (rgb[channelIdx[2]] - inputMean[2]) / inputStd[2];
-            }
-        } else if (channels == 1) {
-            int[] pixels=new int[width*height];
-            scaleImage.getPixels(pixels,0,scaleImage.getWidth(),0,0,scaleImage.getWidth(),scaleImage.getHeight());
-            for (int i = 0; i < pixels.length; i++) {
-                int color = pixels[i];
-                float gray = (float) (red(color) + green(color) + blue(color)) / 3.0f / 255.0f;
-                inputData[i] = (gray - inputMean[0]) / inputStd[0];
-            }
-        } else {
-            Log.i(TAG, "Unsupported channel size " + Integer.toString(channels) + ",  only channel 1 and 3 is " +
-                    "supported!");
-            return false;
-        }
-        float[] pixels = inputData;
-        Log.i(TAG, "pixels " + pixels[0] + " " + pixels[1] + " " + pixels[2] + " " + pixels[3]
-                + " " + pixels[pixels.length / 2] + " " + pixels[pixels.length / 2 + 1] + " " + pixels[pixels.length - 2] + " " + pixels[pixels.length - 1]);
-        Date end = new Date();
-        preprocessTime = (float) (end.getTime() - start.getTime());
-
         // Warm up
         for (int i = 0; i < warmupIterNum; i++) {
-            paddlePredictor.runImage(inputData, width, height, channels, inputImage);
+            paddlePredictor.runImage(inputImage, detLongSize, run_det, run_cls, run_rec);
         }
         warmupIterNum = 0; // do not need warm
         // Run inference
-        start = new Date();
-        ArrayList<OcrResultModel> results = paddlePredictor.runImage(inputData, width, height, channels, inputImage);
-        end = new Date();
+        Date start = new Date();
+        ArrayList<OcrResultModel> results = paddlePredictor.runImage(inputImage, detLongSize, run_det, run_cls, run_rec);
+        Date end = new Date();
         inferenceTime = (end.getTime() - start.getTime()) / (float) inferIterNum;
 
         results = postprocess(results);
-        Log.i(TAG, "[stat] Preprocess Time: " + preprocessTime
-                + " ; Inference Time: " + inferenceTime + " ;Box Size " + results.size());
+        Log.i(TAG, "[stat] Inference Time: " + inferenceTime + " ;Box Size " + results.size());
         drawResults(results);
 
         return true;
     }
 
-
     public boolean isLoaded() {
         return paddlePredictor != null && isLoaded;
     }
@@ -282,10 +195,6 @@ public class Predictor {
         return outputResult;
     }
 
-    public float preprocessTime() {
-        return preprocessTime;
-    }
-
     public float postprocessTime() {
         return postprocessTime;
     }
@@ -310,6 +219,7 @@ public class Predictor {
                 }
             }
             r.setLabel(word.toString());
+            r.setClsLabel(r.getClsIdx() == 1 ? "180" : "0");
         }
         return results;
     }
@@ -319,14 +229,22 @@ public class Predictor {
         for (int i = 0; i < results.size(); i++) {
             OcrResultModel result = results.get(i);
             StringBuilder sb = new StringBuilder("");
-            sb.append(result.getLabel());
-            sb.append(" ").append(result.getConfidence());
-            sb.append("; Points: ");
-            for (Point p : result.getPoints()) {
-                sb.append("(").append(p.x).append(",").append(p.y).append(") ");
+            if(result.getPoints().size()>0){
+                sb.append("Det: ");
+                for (Point p : result.getPoints()) {
+                    sb.append("(").append(p.x).append(",").append(p.y).append(") ");
+                }
+            }
+            if(result.getLabel().length() > 0){
+                sb.append("\n Rec: ").append(result.getLabel());
+                sb.append(",").append(result.getConfidence());
+            }
+            if(result.getClsIdx()!=-1){
+                sb.append(" Cls: ").append(result.getClsLabel());
+                sb.append(",").append(result.getClsConfidence());
             }
             Log.i(TAG, sb.toString()); // show LOG in Logcat panel
-            outputResultSb.append(i + 1).append(": ").append(result.getLabel()).append("\n");
+            outputResultSb.append(i + 1).append(": ").append(sb.toString()).append("\n");
         }
         outputResult = outputResultSb.toString();
         outputImage = inputImage;
@@ -344,6 +262,9 @@ public class Predictor {
         for (OcrResultModel result : results) {
             Path path = new Path();
             List<Point> points = result.getPoints();
+            if(points.size()==0){
+                continue;
+            }
             path.moveTo(points.get(0).x, points.get(0).y);
             for (int i = points.size() - 1; i >= 0; i--) {
                 Point p = points.get(i);

deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/SettingsActivity.java

@@ -20,16 +20,13 @@ public class SettingsActivity extends AppCompatPreferenceActivity implements Sha
     ListPreference etImagePath = null;
     ListPreference lpCPUThreadNum = null;
     ListPreference lpCPUPowerMode = null;
-    ListPreference lpInputColorFormat = null;
-    EditTextPreference etInputShape = null;
-    EditTextPreference etInputMean = null;
-    EditTextPreference etInputStd = null;
+    EditTextPreference etDetLongSize = null;
     EditTextPreference etScoreThreshold = null;
 
     List<String> preInstalledModelPaths = null;
     List<String> preInstalledLabelPaths = null;
     List<String> preInstalledImagePaths = null;
-    List<String> preInstalledInputShapes = null;
+    List<String> preInstalledDetLongSizes = null;
     List<String> preInstalledCPUThreadNums = null;
     List<String> preInstalledCPUPowerModes = null;
     List<String> preInstalledInputColorFormats = null;
@@ -50,7 +47,7 @@ public class SettingsActivity extends AppCompatPreferenceActivity implements Sha
         preInstalledModelPaths = new ArrayList<String>();
         preInstalledLabelPaths = new ArrayList<String>();
         preInstalledImagePaths = new ArrayList<String>();
-        preInstalledInputShapes = new ArrayList<String>();
+        preInstalledDetLongSizes = new ArrayList<String>();
         preInstalledCPUThreadNums = new ArrayList<String>();
         preInstalledCPUPowerModes = new ArrayList<String>();
         preInstalledInputColorFormats = new ArrayList<String>();
@@ -63,10 +60,7 @@ public class SettingsActivity extends AppCompatPreferenceActivity implements Sha
         preInstalledImagePaths.add(getString(R.string.IMAGE_PATH_DEFAULT));
         preInstalledCPUThreadNums.add(getString(R.string.CPU_THREAD_NUM_DEFAULT));
         preInstalledCPUPowerModes.add(getString(R.string.CPU_POWER_MODE_DEFAULT));
-        preInstalledInputColorFormats.add(getString(R.string.INPUT_COLOR_FORMAT_DEFAULT));
-        preInstalledInputShapes.add(getString(R.string.INPUT_SHAPE_DEFAULT));
-        preInstalledInputMeans.add(getString(R.string.INPUT_MEAN_DEFAULT));
-        preInstalledInputStds.add(getString(R.string.INPUT_STD_DEFAULT));
+        preInstalledDetLongSizes.add(getString(R.string.DET_LONG_SIZE_DEFAULT));
         preInstalledScoreThresholds.add(getString(R.string.SCORE_THRESHOLD_DEFAULT));
 
         // Setup UI components
@@ -89,11 +83,7 @@ public class SettingsActivity extends AppCompatPreferenceActivity implements Sha
                 (ListPreference) findPreference(getString(R.string.CPU_THREAD_NUM_KEY));
         lpCPUPowerMode =
                 (ListPreference) findPreference(getString(R.string.CPU_POWER_MODE_KEY));
-        lpInputColorFormat =
-                (ListPreference) findPreference(getString(R.string.INPUT_COLOR_FORMAT_KEY));
-        etInputShape = (EditTextPreference) findPreference(getString(R.string.INPUT_SHAPE_KEY));
-        etInputMean = (EditTextPreference) findPreference(getString(R.string.INPUT_MEAN_KEY));
-        etInputStd = (EditTextPreference) findPreference(getString(R.string.INPUT_STD_KEY));
+        etDetLongSize = (EditTextPreference) findPreference(getString(R.string.DET_LONG_SIZE_KEY));
         etScoreThreshold = (EditTextPreference) findPreference(getString(R.string.SCORE_THRESHOLD_KEY));
     }
 
@@ -112,11 +102,7 @@ public class SettingsActivity extends AppCompatPreferenceActivity implements Sha
                 editor.putString(getString(R.string.IMAGE_PATH_KEY), preInstalledImagePaths.get(modelIdx));
                 editor.putString(getString(R.string.CPU_THREAD_NUM_KEY), preInstalledCPUThreadNums.get(modelIdx));
                 editor.putString(getString(R.string.CPU_POWER_MODE_KEY), preInstalledCPUPowerModes.get(modelIdx));
-                editor.putString(getString(R.string.INPUT_COLOR_FORMAT_KEY),
-                        preInstalledInputColorFormats.get(modelIdx));
-                editor.putString(getString(R.string.INPUT_SHAPE_KEY), preInstalledInputShapes.get(modelIdx));
-                editor.putString(getString(R.string.INPUT_MEAN_KEY), preInstalledInputMeans.get(modelIdx));
-                editor.putString(getString(R.string.INPUT_STD_KEY), preInstalledInputStds.get(modelIdx));
+                editor.putString(getString(R.string.DET_LONG_SIZE_KEY), preInstalledDetLongSizes.get(modelIdx));
                 editor.putString(getString(R.string.SCORE_THRESHOLD_KEY),
                         preInstalledScoreThresholds.get(modelIdx));
                 editor.apply();
@@ -129,10 +115,7 @@ public class SettingsActivity extends AppCompatPreferenceActivity implements Sha
         etImagePath.setEnabled(enableCustomSettings);
         lpCPUThreadNum.setEnabled(enableCustomSettings);
         lpCPUPowerMode.setEnabled(enableCustomSettings);
-        lpInputColorFormat.setEnabled(enableCustomSettings);
-        etInputShape.setEnabled(enableCustomSettings);
-        etInputMean.setEnabled(enableCustomSettings);
-        etInputStd.setEnabled(enableCustomSettings);
+        etDetLongSize.setEnabled(enableCustomSettings);
         etScoreThreshold.setEnabled(enableCustomSettings);
         modelPath = sharedPreferences.getString(getString(R.string.MODEL_PATH_KEY),
                 getString(R.string.MODEL_PATH_DEFAULT));
@@ -144,14 +127,8 @@ public class SettingsActivity extends AppCompatPreferenceActivity implements Sha
                 getString(R.string.CPU_THREAD_NUM_DEFAULT));
         String cpuPowerMode = sharedPreferences.getString(getString(R.string.CPU_POWER_MODE_KEY),
                 getString(R.string.CPU_POWER_MODE_DEFAULT));
-        String inputColorFormat = sharedPreferences.getString(getString(R.string.INPUT_COLOR_FORMAT_KEY),
-                getString(R.string.INPUT_COLOR_FORMAT_DEFAULT));
-        String inputShape = sharedPreferences.getString(getString(R.string.INPUT_SHAPE_KEY),
-                getString(R.string.INPUT_SHAPE_DEFAULT));
-        String inputMean = sharedPreferences.getString(getString(R.string.INPUT_MEAN_KEY),
-                getString(R.string.INPUT_MEAN_DEFAULT));
-        String inputStd = sharedPreferences.getString(getString(R.string.INPUT_STD_KEY),
-                getString(R.string.INPUT_STD_DEFAULT));
+        String detLongSize = sharedPreferences.getString(getString(R.string.DET_LONG_SIZE_KEY),
+                getString(R.string.DET_LONG_SIZE_DEFAULT));
         String scoreThreshold = sharedPreferences.getString(getString(R.string.SCORE_THRESHOLD_KEY),
                 getString(R.string.SCORE_THRESHOLD_DEFAULT));
         etModelPath.setSummary(modelPath);
@@ -164,14 +141,8 @@ public class SettingsActivity extends AppCompatPreferenceActivity implements Sha
         lpCPUThreadNum.setSummary(cpuThreadNum);
         lpCPUPowerMode.setValue(cpuPowerMode);
         lpCPUPowerMode.setSummary(cpuPowerMode);
-        lpInputColorFormat.setValue(inputColorFormat);
-        lpInputColorFormat.setSummary(inputColorFormat);
-        etInputShape.setSummary(inputShape);
-        etInputShape.setText(inputShape);
-        etInputMean.setSummary(inputMean);
-        etInputMean.setText(inputMean);
-        etInputStd.setSummary(inputStd);
-        etInputStd.setText(inputStd);
+        etDetLongSize.setSummary(detLongSize);
+        etDetLongSize.setText(detLongSize);
         etScoreThreshold.setText(scoreThreshold);
         etScoreThreshold.setSummary(scoreThreshold);
     }

deploy/android_demo/app/src/main/res/layout/activity_main.xml

@@ -23,13 +23,7 @@
                 android:layout_height="wrap_content"
                 android:orientation="horizontal">
 
-                <Button
-                    android:id="@+id/btn_load_model"
-                    android:layout_width="0dp"
-                    android:layout_height="wrap_content"
-                    android:layout_weight="1"
-                    android:onClick="btn_load_model_click"
-                    android:text="加载模型" />
+
                 <Button
                     android:id="@+id/btn_run_model"
                     android:layout_width="0dp"
@@ -52,7 +46,45 @@
                     android:onClick="btn_choice_img_click"
                     android:text="选取图片" />
 
+                <Button
+                    android:id="@+id/btn_reset_img"
+                    android:layout_width="0dp"
+                    android:layout_height="wrap_content"
+                    android:layout_weight="1"
+                    android:onClick="btn_reset_img_click"
+                    android:text="清空绘图" />
+
             </LinearLayout>
+            <LinearLayout
+                android:id="@+id/run_mode_layout"
+                android:layout_width="fill_parent"
+                android:layout_height="wrap_content"
+                android:orientation="horizontal">
+
+                <CheckBox
+                    android:id="@+id/cb_opencl"
+                    android:layout_width="0dp"
+                    android:layout_weight="1"
+                    android:layout_height="wrap_content"
+                    android:text="开启OPENCL"
+                    android:onClick="cb_opencl_click"
+                    android:visibility="gone"/>
+                <TextView
+                    android:layout_width="0dp"
+                    android:layout_weight="0.5"
+                    android:layout_height="wrap_content"
+                    android:text="运行模式:"/>
+
+                <Spinner
+                    android:id="@+id/sp_run_mode"
+                    android:layout_width="0dp"
+                    android:layout_weight="1.5"
+                    android:layout_height="wrap_content"
+                    android:entries="@array/run_Model"
+                    />
+
+            </LinearLayout>
+
             <TextView
                     android:id="@+id/tv_input_setting"
                     android:layout_width="wrap_content"
@@ -60,7 +92,7 @@
                     android:scrollbars="vertical"
                     android:layout_marginLeft="12dp"
                     android:layout_marginRight="12dp"
-                    android:layout_marginTop="10dp"
+                    android:layout_marginTop="5dp"
                     android:layout_marginBottom="5dp"
                     android:lineSpacingExtra="4dp"
                     android:singleLine="false"

deploy/android_demo/app/src/main/res/layout/activity_mini.xml

-<?xml version="1.0" encoding="utf-8"?>
-<!-- for MiniActivity Use Only -->
-<androidx.constraintlayout.widget.ConstraintLayout xmlns:android="http://schemas.android.com/apk/res/android"
-    xmlns:app="http://schemas.android.com/apk/res-auto"
-    xmlns:tools="http://schemas.android.com/tools"
-    android:layout_width="match_parent"
-    android:layout_height="match_parent"
-    app:layout_constraintLeft_toLeftOf="parent"
-    app:layout_constraintLeft_toRightOf="parent"
-    tools:context=".MainActivity">
-
-    <TextView
-        android:id="@+id/sample_text"
-        android:layout_width="0dp"
-        android:layout_height="wrap_content"
-        android:text="Hello World!"
-        app:layout_constraintLeft_toLeftOf="parent"
-        app:layout_constraintRight_toRightOf="parent"
-        app:layout_constraintTop_toBottomOf="@id/imageView"
-        android:scrollbars="vertical"
-        />
-
-    <ImageView
-        android:id="@+id/imageView"
-        android:layout_width="wrap_content"
-        android:layout_height="wrap_content"
-        android:paddingTop="20dp"
-        android:paddingBottom="20dp"
-        app:layout_constraintBottom_toTopOf="@id/imageView"
-        app:layout_constraintLeft_toLeftOf="parent"
-        app:layout_constraintRight_toRightOf="parent"
-        app:layout_constraintTop_toTopOf="parent"
-        tools:srcCompat="@tools:sample/avatars" />
-
-    <Button
-        android:id="@+id/button"
-        android:layout_width="wrap_content"
-        android:layout_height="wrap_content"
-        android:layout_marginBottom="4dp"
-        android:text="Button"
-        app:layout_constraintBottom_toBottomOf="parent"
-        app:layout_constraintLeft_toLeftOf="parent"
-        app:layout_constraintRight_toRightOf="parent"
-        tools:layout_editor_absoluteX="161dp" />
-
-</androidx.constraintlayout.widget.ConstraintLayout>
\ No newline at end of file

deploy/android_demo/app/src/main/res/values/arrays.xml

 <?xml version="1.0" encoding="utf-8"?>
 <resources>
     <string-array name="image_name_entries">
-        <item>0.jpg</item>
-        <item>90.jpg</item>
-        <item>180.jpg</item>
-        <item>270.jpg</item>
+        <item>det_0.jpg</item>
+        <item>det_90.jpg</item>
+        <item>det_180.jpg</item>
+        <item>det_270.jpg</item>
+        <item>rec_0.jpg</item>
+        <item>rec_0_180.jpg</item>
+        <item>rec_1.jpg</item>
+        <item>rec_1_180.jpg</item>
     </string-array>
     <string-array name="image_name_values">
-        <item>images/0.jpg</item>
-        <item>images/90.jpg</item>
-        <item>images/180.jpg</item>
-        <item>images/270.jpg</item>
+        <item>images/det_0.jpg</item>
+        <item>images/det_90.jpg</item>
+        <item>images/det_180.jpg</item>
+        <item>images/det_270.jpg</item>
+        <item>images/rec_0.jpg</item>
+        <item>images/rec_0_180.jpg</item>
+        <item>images/rec_1.jpg</item>
+        <item>images/rec_1_180.jpg</item>
     </string-array>
     <string-array name="cpu_thread_num_entries">
         <item>1 threads</item>
@@ -48,4 +56,12 @@
         <item>BGR</item>
         <item>RGB</item>
     </string-array>
+    <string-array name="run_Model">
+        <item>检测+分类+识别</item>
+        <item>检测+识别</item>
+        <item>分类+识别</item>
+        <item>检测</item>
+        <item>识别</item>
+        <item>分类</item>
+    </string-array>
 </resources>
\ No newline at end of file

deploy/android_demo/app/src/main/res/values/strings.xml

 <resources>
-    <string name="app_name">OCR Chinese</string>
+    <string name="app_name">PaddleOCR</string>
     <string name="CHOOSE_PRE_INSTALLED_MODEL_KEY">CHOOSE_PRE_INSTALLED_MODEL_KEY</string>
     <string name="ENABLE_CUSTOM_SETTINGS_KEY">ENABLE_CUSTOM_SETTINGS_KEY</string>
     <string name="MODEL_PATH_KEY">MODEL_PATH_KEY</string>
@@ -7,20 +7,14 @@
     <string name="IMAGE_PATH_KEY">IMAGE_PATH_KEY</string>
     <string name="CPU_THREAD_NUM_KEY">CPU_THREAD_NUM_KEY</string>
     <string name="CPU_POWER_MODE_KEY">CPU_POWER_MODE_KEY</string>
-    <string name="INPUT_COLOR_FORMAT_KEY">INPUT_COLOR_FORMAT_KEY</string>
-    <string name="INPUT_SHAPE_KEY">INPUT_SHAPE_KEY</string>
-    <string name="INPUT_MEAN_KEY">INPUT_MEAN_KEY</string>
-    <string name="INPUT_STD_KEY">INPUT_STD_KEY</string>
+    <string name="DET_LONG_SIZE_KEY">DET_LONG_SIZE_KEY</string>
     <string name="SCORE_THRESHOLD_KEY">SCORE_THRESHOLD_KEY</string>
-    <string name="MODEL_PATH_DEFAULT">models/ocr_v2_for_cpu</string>
+    <string name="MODEL_PATH_DEFAULT">models/ch_PP-OCRv2</string>
     <string name="LABEL_PATH_DEFAULT">labels/ppocr_keys_v1.txt</string>
-    <string name="IMAGE_PATH_DEFAULT">images/0.jpg</string>
+    <string name="IMAGE_PATH_DEFAULT">images/det_0.jpg</string>
     <string name="CPU_THREAD_NUM_DEFAULT">4</string>
     <string name="CPU_POWER_MODE_DEFAULT">LITE_POWER_HIGH</string>
-    <string name="INPUT_COLOR_FORMAT_DEFAULT">BGR</string>
-    <string name="INPUT_SHAPE_DEFAULT">1,3,960</string>
-    <string name="INPUT_MEAN_DEFAULT">0.485, 0.456, 0.406</string>
-    <string name="INPUT_STD_DEFAULT">0.229,0.224,0.225</string>
+    <string name="DET_LONG_SIZE_DEFAULT">960</string>
     <string name="SCORE_THRESHOLD_DEFAULT">0.1</string>
 </resources>
 

deploy/android_demo/app/src/main/res/xml/settings.xml

@@ -47,26 +47,10 @@
                 android:entryValues="@array/cpu_power_mode_values"/>
     </PreferenceCategory>
     <PreferenceCategory android:title="Input Settings">
-        <ListPreference
-                android:defaultValue="@string/INPUT_COLOR_FORMAT_DEFAULT"
-                android:key="@string/INPUT_COLOR_FORMAT_KEY"
-                android:negativeButtonText="@null"
-                android:positiveButtonText="@null"
-                android:title="Input Color Format: BGR or RGB"
-                android:entries="@array/input_color_format_entries"
-                android:entryValues="@array/input_color_format_values"/>
-        <EditTextPreference
-                android:key="@string/INPUT_SHAPE_KEY"
-                android:defaultValue="@string/INPUT_SHAPE_DEFAULT"
-                android:title="Input Shape: (1,1,max_width_height) or (1,3,max_width_height)" />
-        <EditTextPreference
-                android:key="@string/INPUT_MEAN_KEY"
-                android:defaultValue="@string/INPUT_MEAN_DEFAULT"
-                android:title="Input Mean: (channel/255-mean)/std" />
         <EditTextPreference
-                android:key="@string/INPUT_STD_KEY"
-                android:defaultValue="@string/INPUT_STD_DEFAULT"
-                android:title="Input Std: (channel/255-mean)/std" />
+                android:key="@string/DET_LONG_SIZE_KEY"
+                android:defaultValue="@string/DET_LONG_SIZE_DEFAULT"
+                android:title="det long size" />
     </PreferenceCategory>
     <PreferenceCategory android:title="Output Settings">
         <EditTextPreference

deploy/lite/readme.md

+- [端侧部署](#端侧部署)
+  - [1. 准备环境](#1-准备环境)
+    - [运行准备](#运行准备)
+    - [1.1 准备交叉编译环境](#11-准备交叉编译环境)
+    - [1.2 准备预测库](#12-准备预测库)
+  - [2 开始运行](#2-开始运行)
+    - [2.1 模型优化](#21-模型优化)
+    - [2.2 与手机联调](#22-与手机联调)
+      - [注意：](#注意)
+  - [FAQ](#faq)
+  
 # 端侧部署
 
 本教程将介绍基于[Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite) 在移动端部署PaddleOCR超轻量中文检测、识别模型的详细步骤。
@@ -26,17 +37,17 @@ Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理
 
       | 平台 | 预测库下载链接 |
       |---|---|
-      |Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.9/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.9/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
-      |IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.9/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.9/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
+      |Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
+      |IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
 
-      注：1. 上述预测库为PaddleLite 2.9分支编译得到，有关PaddleLite 2.9 详细信息可参考 [链接](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.9) 。
+      注：1. 上述预测库为PaddleLite 2.10分支编译得到，有关PaddleLite 2.10 详细信息可参考 [链接](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10) 。
 
 - 2. [推荐]编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
 ```
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
-# 切换到Paddle-Lite release/v2.9 稳定分支
-git checkout release/v2.9
+# 切换到Paddle-Lite release/v2.10 稳定分支
+git checkout release/v2.10
 ./lite/tools/build_android.sh  --arch=armv8  --with_cv=ON --with_extra=ON
 ```
 
@@ -85,8 +96,8 @@ Paddle-Lite 提供了多种策略来自动优化原始的模型，其中包括
 
 |模型版本|模型简介|模型大小|检测模型|文本方向分类模型|识别模型|Paddle-Lite版本|
 |---|---|---|---|---|---|---|
-|V2.0|超轻量中文OCR 移动端模型|7.8M|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_opt.nb)|v2.9|
-|V2.0(slim)|超轻量中文OCR 移动端模型|3.3M|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_slim_opt.nb)|v2.9|
+|PP-OCRv2|蒸馏版超轻量中文OCR移动端模型|11M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)|v2.10|
+|PP-OCRv2(slim)|蒸馏版超轻量中文OCR移动端模型|4.6M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)|v2.10|
 
 如果直接使用上述表格中的模型进行部署，可略过下述步骤，直接阅读 [2.2节](#2.2与手机联调)。
 
@@ -97,7 +108,7 @@ Paddle-Lite 提供了多种策略来自动优化原始的模型，其中包括
 # 如果准备环境时已经clone了Paddle-Lite，则不用重新clone Paddle-Lite
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
-git checkout release/v2.9
+git checkout release/v2.10
 # 启动编译
 ./lite/tools/build.sh build_optimize_tool
 ```
@@ -123,15 +134,15 @@ cd build.opt/lite/api/
 下面以PaddleOCR的超轻量中文模型为例，介绍使用编译好的opt文件完成inference模型到Paddle-Lite优化模型的转换。
 
 ```
-# 【推荐】 下载PaddleOCR V2.0版本的中英文 inference模型
-wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_det_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_det_slim_infer.tar
-wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_rec_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_rec_slim_infer.tar
+# 【推荐】 下载 PP-OCRv2版本的中英文 inference模型
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_slim_quant_infer.tar && tar xf  ch_PP-OCRv2_det_slim_quant_infer.tar
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_quant_infer.tar && tar xf  ch_PP-OCRv2_rec_slim_quant_infer.tar
 wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_cls_slim_infer.tar
-# 转换V2.0检测模型
-./opt --model_file=./ch_ppocr_mobile_v2.0_det_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_det_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_det_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
-# 转换V2.0识别模型
-./opt --model_file=./ch_ppocr_mobile_v2.0_rec_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_rec_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_rec_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
-# 转换V2.0方向分类器模型
+# 转换检测模型
+./opt --model_file=./ch_PP-OCRv2_det_slim_quant_infer/inference.pdmodel  --param_file=./ch_PP-OCRv2_det_slim_quant_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv2_det_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# 转换识别模型
+./opt --model_file=./ch_PP-OCRv2_rec_slim_quant_infer/inference.pdmodel  --param_file=./ch_PP-OCRv2_rec_slim_quant_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv2_rec_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# 转换方向分类器模型
 ./opt --model_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_cls_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
 
 ```
@@ -186,15 +197,15 @@ wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls
  ```
 
  准备测试图像，以`PaddleOCR/doc/imgs/11.jpg`为例，将测试的图像复制到`demo/cxx/ocr/debug/`文件夹下。
- 准备lite opt工具优化后的模型文件，比如使用`ch_ppocr_mobile_v2.0_det_slim_opt.nb，ch_ppocr_mobile_v2.0_rec_slim_opt.nb, ch_ppocr_mobile_v2.0_cls_slim_opt.nb`，模型文件放置在`demo/cxx/ocr/debug/`文件夹下。
+ 准备lite opt工具优化后的模型文件，比如使用`ch_PP-OCRv2_det_slim_opt.ch_PP-OCRv2_rec_slim_rec.nb, ch_ppocr_mobile_v2.0_cls_slim_opt.nb`，模型文件放置在`demo/cxx/ocr/debug/`文件夹下。
 
  执行完成后，ocr文件夹下将有如下文件格式：
 
 ```
 demo/cxx/ocr/
 |-- debug/  
-|   |--ch_ppocr_mobile_v2.0_det_slim_opt.nb           优化后的检测模型文件
-|   |--ch_ppocr_mobile_v2.0_rec_slim_opt.nb           优化后的识别模型文件
+|   |--ch_PP-OCRv2_det_slim_opt.nb           优化后的检测模型文件
+|   |--ch_PP-OCRv2_rec_slim_opt.nb           优化后的识别模型文件
 |   |--ch_ppocr_mobile_v2.0_cls_slim_opt.nb           优化后的文字方向分类器模型文件
 |   |--11.jpg                           待测试图像
 |   |--ppocr_keys_v1.txt                中文字典文件
@@ -250,7 +261,7 @@ use_direction_classify  0  # 是否使用方向分类器，0表示不使用，1
  export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
  # 开始使用，ocr_db_crnn可执行文件的使用方式为:
  # ./ocr_db_crnn  检测模型文件 方向分类器模型文件  识别模型文件  测试图像路径  字典文件路径
- ./ocr_db_crnn ch_ppocr_mobile_v2.0_det_slim_opt.nb  ch_ppocr_mobile_v2.0_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  ./11.jpg  ppocr_keys_v1.txt
+ ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  ./11.jpg  ppocr_keys_v1.txt
  ```
 
  如果对代码做了修改，则需要重新编译并push到手机上。

deploy/lite/readme_en.md

+- [Tutorial of PaddleOCR Mobile deployment](#tutorial-of-paddleocr-mobile-deployment)
+  - [1. Preparation](#1-preparation)
+    - [Preparation environment](#preparation-environment)
+    - [1.1 Prepare the cross-compilation environment](#11-prepare-the-cross-compilation-environment)
+    - [1.2 Prepare Paddle-Lite library](#12-prepare-paddle-lite-library)
+  - [2 Run](#2-run)
+    - [2.1 Inference Model Optimization](#21-inference-model-optimization)
+    - [2.2 Run optimized model on Phone](#22-run-optimized-model-on-phone)
+      - [注意：](#注意)
+  - [FAQ](#faq)
+
 # Tutorial of PaddleOCR Mobile deployment
 
 This tutorial will introduce how to use [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite) to deploy PaddleOCR ultra-lightweight Chinese and English detection models on mobile phones.
@@ -28,17 +39,17 @@ There are two ways to obtain the Paddle-Lite library：
 
       | Platform | Paddle-Lite library download link |
       |---|---|
-      |Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.9/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.9/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
-      |IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.9/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.9/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
+      |Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
+      |IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
 
-      Note: 1. The above Paddle-Lite library is compiled from the Paddle-Lite 2.9 branch. For more information about Paddle-Lite 2.9, please refer to [link](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.9).
+      Note: 1. The above Paddle-Lite library is compiled from the Paddle-Lite 2.10 branch. For more information about Paddle-Lite 2.10, please refer to [link](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10).
 
 - 2. [Recommended] Compile Paddle-Lite to get the prediction library. The compilation method of Paddle-Lite is as follows：
 ```
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
-# Switch to Paddle-Lite release/v2.8 stable branch
-git checkout release/v2.8
+# Switch to Paddle-Lite release/v2.10 stable branch
+git checkout release/v2.10
 ./lite/tools/build_android.sh  --arch=armv8  --with_cv=ON --with_extra=ON
 ```
 
@@ -87,10 +98,10 @@ The following table also provides a series of models that can be deployed on mob
 
 |Version|Introduction|Model size|Detection model|Text Direction model|Recognition model|Paddle-Lite branch|
 |---|---|---|---|---|---|---|
-|V2.0|extra-lightweight chinese OCR optimized model|7.8M|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_opt.nb)|v2.9|
-|V2.0(slim)|extra-lightweight chinese OCR optimized model|3.3M|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_slim_opt.nb)|v2.9|
+|PP-OCRv2|extra-lightweight chinese OCR optimized model|11M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)|v2.10|
+|PP-OCRv2(slim)|extra-lightweight chinese OCR optimized model|4.6M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)|v2.10|
 
-If you directly use the model in the above table for deployment, you can skip the following steps and directly read [Section 2.2](#2.2 Run optimized model on Phone).
+If you directly use the model in the above table for deployment, you can skip the following steps and directly read [Section 2.2](#2.2-Run-optimized-model-on-Phone).
 
 If the model to be deployed is not in the above table, you need to follow the steps below to obtain the optimized model.
 
@@ -98,7 +109,7 @@ The `opt` tool can be obtained by compiling Paddle Lite.
 ```
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
-git checkout release/v2.9
+git checkout release/v2.10
 ./lite/tools/build.sh build_optimize_tool
 ```
 
@@ -124,22 +135,22 @@ cd build.opt/lite/api/
 The following takes the ultra-lightweight Chinese model of PaddleOCR as an example to introduce the use of the compiled opt file to complete the conversion of the inference model to the Paddle-Lite optimized model
 
 ```
-# [Recommendation] Download the Chinese and English inference model of PaddleOCR V2.0
-wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_det_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_det_slim_infer.tar
-wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_rec_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_rec_slim_infer.tar
+# 【[Recommendation] Download the Chinese and English inference model of PP-OCRv2
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_slim_quant_infer.tar && tar xf  ch_PP-OCRv2_det_slim_quant_infer.tar
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_quant_infer.tar && tar xf  ch_PP-OCRv2_rec_slim_quant_infer.tar
 wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_cls_slim_infer.tar
-# Convert V2.0 detection model
-./opt --model_file=./ch_ppocr_mobile_v2.0_det_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_det_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_det_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
-# Convert V2.0 recognition model
-./opt --model_file=./ch_ppocr_mobile_v2.0_rec_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_rec_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_rec_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
-# Convert V2.0 angle classifier model
+# Convert detection model
+./opt --model_file=./ch_PP-OCRv2_det_slim_quant_infer/inference.pdmodel  --param_file=./ch_PP-OCRv2_det_slim_quant_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv2_det_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# Convert recognition model
+./opt --model_file=./ch_PP-OCRv2_rec_slim_quant_infer/inference.pdmodel  --param_file=./ch_PP-OCRv2_rec_slim_quant_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv2_rec_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# Convert angle classifier model
 ./opt --model_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_cls_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
 
 ```
 
 After the conversion is successful, there will be more files ending with `.nb` in the inference model directory, which is the successfully converted model file.
 
-<a name="2.2 Run optimized model on Phone"></a>
+<a name="2.2-Run-optimized-model-on-Phone"></a>
 ### 2.2 Run optimized model on Phone
 
 Some preparatory work is required first.
@@ -194,8 +205,8 @@ The structure of the OCR demo is as follows after the above command is executed:
 ```
 demo/cxx/ocr/
 |-- debug/  
-|   |--ch_ppocr_mobile_v2.0_det_slim_opt.nb           Detection model
-|   |--ch_ppocr_mobile_v2.0_rec_slim_opt.nb           Recognition model
+|   |--ch_PP-OCRv2_det_slim_opt.nb           Detection model
+|   |--ch_PP-OCRv2_rec_slim_opt.nb           Recognition model
 |   |--ch_ppocr_mobile_v2.0_cls_slim_opt.nb           Text direction classification model
 |   |--11.jpg                           Image for OCR
 |   |--ppocr_keys_v1.txt                Dictionary file
@@ -249,7 +260,7 @@ After the above steps are completed, you can use adb to push the file to the pho
  export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
  # The use of ocr_db_crnn is:
  # ./ocr_db_crnn Detection model file Orientation classifier model file Recognition model file Test image path Dictionary file path
- ./ocr_db_crnn ch_ppocr_mobile_v2.0_det_opt.nb  ch_ppocr_mobile_v2.0_rec_opt.nb  ch_ppocr_mobile_v2.0_cls_opt.nb  ./11.jpg  ppocr_keys_v1.txt
+ ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_opt.nb  ./11.jpg  ppocr_keys_v1.txt
  ```
 
 If you modify the code, you need to recompile and push to the phone.

doc/doc_ch/models_list.md

@@ -6,13 +6,14 @@
 > 3. 本文档提供的是PPOCR自研模型列表，更多基于公开数据集的算法介绍与预训练模型可以参考：[算法概览文档](./algorithm_overview.md)。
 
 
-- [1. 文本检测模型](#文本检测模型)
-- [2. 文本识别模型](#文本识别模型)
-    - [2.1 中文识别模型](#中文识别模型)
-    - [2.2 英文识别模型](#英文识别模型)
-    - [2.3 多语言识别模型](#多语言识别模型)
-- [3. 文本方向分类模型](#文本方向分类模型)
-- [4. Paddle-Lite 模型](#Paddle-Lite模型)
+- [PP-OCR系列模型列表（V2.1，2021年9月6日更新）](#pp-ocr系列模型列表v212021年9月6日更新)
+  - [1. 文本检测模型](#1-文本检测模型)
+  - [2. 文本识别模型](#2-文本识别模型)
+    - [2.1 中文识别模型](#21-中文识别模型)
+    - [2.2 英文识别模型](#22-英文识别模型)
+    - [2.3 多语言识别模型（更多语言持续更新中...）](#23-多语言识别模型更多语言持续更新中)
+  - [3. 文本方向分类模型](#3-文本方向分类模型)
+  - [4. Paddle-Lite 模型](#4-paddle-lite-模型)
 
 PaddleOCR提供的可下载模型包括`推理模型`、`训练模型`、`预训练模型`、`slim模型`，模型区别说明如下：
 
@@ -100,6 +101,8 @@ PaddleOCR提供的可下载模型包括`推理模型`、`训练模型`、`预训
 
 |模型版本|模型简介|模型大小|检测模型|文本方向分类模型|识别模型|Paddle-Lite版本|
 |---|---|---|---|---|---|---|
+|PP-OCRv2|蒸馏版超轻量中文OCR移动端模型|11M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)|v2.10|
+|PP-OCRv2(slim)|蒸馏版超轻量中文OCR移动端模型|4.6M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)|v2.10|
 |PP-OCRv2|蒸馏版超轻量中文OCR移动端模型|11M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer_opt.nb)|v2.9|
 |PP-OCRv2(slim)|蒸馏版超轻量中文OCR移动端模型|4.9M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_opt.nb)|v2.9|
 |V2.0|ppocr_v2.0超轻量中文OCR移动端模型|7.8M|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_opt.nb)|v2.9|

doc/doc_en/models_list_en.md

@@ -94,6 +94,8 @@ For more supported languages, please refer to : [Multi-language model](./multi_l
 ## 4. Paddle-Lite Model
 |Version|Introduction|Model size|Detection model|Text Direction model|Recognition model|Paddle-Lite branch|
 |---|---|---|---|---|---|---|
+|PP-OCRv2|extra-lightweight chinese OCR optimized model|11M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)|v2.10|
+|PP-OCRv2(slim)|extra-lightweight chinese OCR optimized model|4.6M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)|v2.10|
 |PP-OCRv2|extra-lightweight chinese OCR optimized model|11M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer_opt.nb)|v2.9|
 |PP-OCRv2(slim)|extra-lightweight chinese OCR optimized model|4.9M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_opt.nb)|v2.9|
 |V2.0|ppocr_v2.0 extra-lightweight chinese OCR optimized model|7.8M|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_opt.nb)|v2.9|