Merge pull request #218 from joey12300/develop

encapsulate GenerateColormap in Visualize

deploy/cpp/demo/classifier.cpp

@@ -62,8 +62,6 @@ int main(int argc, char** argv) {
              FLAGS_use_ir_optim);
 
   // 进行预测
-  double total_running_time_s = 0.0;
-  double total_imread_time_s = 0.0;
   int imgs = 1;
   if (FLAGS_image_list != "") {
     std::ifstream inf(FLAGS_image_list);
@@ -79,7 +77,6 @@ int main(int argc, char** argv) {
     }
     imgs = image_paths.size();
     for (int i = 0; i < image_paths.size(); i += FLAGS_batch_size) {
-      auto start = system_clock::now();
       // 读图像
       int im_vec_size =
           std::min(static_cast<int>(image_paths.size()), i + FLAGS_batch_size);
@@ -91,19 +88,7 @@ int main(int argc, char** argv) {
       for (int j = i; j < im_vec_size; ++j) {
         im_vec[j - i] = std::move(cv::imread(image_paths[j], 1));
       }
-      auto imread_end = system_clock::now();
       model.predict(im_vec, &results, thread_num);
-
-      auto imread_duration = duration_cast<microseconds>(imread_end - start);
-      total_imread_time_s += static_cast<double>(imread_duration.count()) *
-                             microseconds::period::num /
-                             microseconds::period::den;
-
-      auto end = system_clock::now();
-      auto duration = duration_cast<microseconds>(end - start);
-      total_running_time_s += static_cast<double>(duration.count()) *
-                              microseconds::period::num /
-                              microseconds::period::den;
       for (int j = i; j < im_vec_size; ++j) {
         std::cout << "Path:" << image_paths[j]
                   << ", predict label: " << results[j - i].category
@@ -112,23 +97,12 @@ int main(int argc, char** argv) {
       }
     }
   } else {
-    auto start = system_clock::now();
     PaddleX::ClsResult result;
     cv::Mat im = cv::imread(FLAGS_image, 1);
     model.predict(im, &result);
-    auto end = system_clock::now();
-    auto duration = duration_cast<microseconds>(end - start);
-    total_running_time_s += static_cast<double>(duration.count()) *
-                            microseconds::period::num /
-                            microseconds::period::den;
     std::cout << "Predict label: " << result.category
               << ", label_id:" << result.category_id
               << ", score: " << result.score << std::endl;
   }
-  std::cout << "Total running time: " << total_running_time_s
-            << " s, average running time: " << total_running_time_s / imgs
-            << " s/img, total read img time: " << total_imread_time_s
-            << " s, average read time: " << total_imread_time_s / imgs
-            << " s/img, batch_size = " << FLAGS_batch_size << std::endl;
   return 0;
 }

deploy/cpp/demo/detector.cpp

@@ -65,11 +65,7 @@ int main(int argc, char** argv) {
              FLAGS_gpu_id,
              FLAGS_key,
              FLAGS_use_ir_optim);
-
-  double total_running_time_s = 0.0;
-  double total_imread_time_s = 0.0;
   int imgs = 1;
-  auto colormap = PaddleX::GenerateColorMap(model.labels.size());
   std::string save_dir = "output";
   // 进行预测
   if (FLAGS_image_list != "") {
@@ -85,7 +81,6 @@ int main(int argc, char** argv) {
     }
     imgs = image_paths.size();
     for (int i = 0; i < image_paths.size(); i += FLAGS_batch_size) {
-      auto start = system_clock::now();
       int im_vec_size =
           std::min(static_cast<int>(image_paths.size()), i + FLAGS_batch_size);
       std::vector<cv::Mat> im_vec(im_vec_size - i);
@@ -96,17 +91,7 @@ int main(int argc, char** argv) {
       for (int j = i; j < im_vec_size; ++j) {
         im_vec[j - i] = std::move(cv::imread(image_paths[j], 1));
       }
-      auto imread_end = system_clock::now();
       model.predict(im_vec, &results, thread_num);
-      auto imread_duration = duration_cast<microseconds>(imread_end - start);
-      total_imread_time_s += static_cast<double>(imread_duration.count()) *
-                             microseconds::period::num /
-                             microseconds::period::den;
-      auto end = system_clock::now();
-      auto duration = duration_cast<microseconds>(end - start);
-      total_running_time_s += static_cast<double>(duration.count()) *
-                              microseconds::period::num /
-                              microseconds::period::den;
       // 输出结果目标框
       for (int j = 0; j < im_vec_size - i; ++j) {
         for (int k = 0; k < results[j].boxes.size(); ++k) {
@@ -124,7 +109,7 @@ int main(int argc, char** argv) {
       // 可视化
       for (int j = 0; j < im_vec_size - i; ++j) {
         cv::Mat vis_img = PaddleX::Visualize(
-            im_vec[j], results[j], model.labels, colormap, FLAGS_threshold);
+            im_vec[j], results[j], model.labels, FLAGS_threshold);
         std::string save_path =
             PaddleX::generate_save_path(FLAGS_save_dir, image_paths[i + j]);
         cv::imwrite(save_path, vis_img);
@@ -132,15 +117,9 @@ int main(int argc, char** argv) {
       }
     }
   } else {
-    auto start = system_clock::now();
     PaddleX::DetResult result;
     cv::Mat im = cv::imread(FLAGS_image, 1);
     model.predict(im, &result);
-    auto end = system_clock::now();
-    auto duration = duration_cast<microseconds>(end - start);
-    total_running_time_s += static_cast<double>(duration.count()) *
-                            microseconds::period::num /
-                            microseconds::period::den;
     // 输出结果目标框
     for (int i = 0; i < result.boxes.size(); ++i) {
       std::cout << "image file: " << FLAGS_image << std::endl;
@@ -155,7 +134,7 @@ int main(int argc, char** argv) {
 
     // 可视化
     cv::Mat vis_img =
-        PaddleX::Visualize(im, result, model.labels, colormap, FLAGS_threshold);
+        PaddleX::Visualize(im, result, model.labels, FLAGS_threshold);
     std::string save_path =
         PaddleX::generate_save_path(FLAGS_save_dir, FLAGS_image);
     cv::imwrite(save_path, vis_img);
@@ -163,11 +142,5 @@ int main(int argc, char** argv) {
     std::cout << "Visualized output saved as " << save_path << std::endl;
   }
 
-  std::cout << "Total running time: " << total_running_time_s
-            << " s, average running time: " << total_running_time_s / imgs
-            << " s/img, total read img time: " << total_imread_time_s
-            << " s, average read img time: " << total_imread_time_s / imgs
-            << " s, batch_size = " << FLAGS_batch_size << std::endl;
-
   return 0;
 }

deploy/cpp/demo/segmenter.cpp

@@ -62,11 +62,7 @@ int main(int argc, char** argv) {
              FLAGS_gpu_id,
              FLAGS_key,
              FLAGS_use_ir_optim);
-
-  double total_running_time_s = 0.0;
-  double total_imread_time_s = 0.0;
   int imgs = 1;
-  auto colormap = PaddleX::GenerateColorMap(model.labels.size());
   // 进行预测
   if (FLAGS_image_list != "") {
     std::ifstream inf(FLAGS_image_list);
@@ -81,7 +77,6 @@ int main(int argc, char** argv) {
     }
     imgs = image_paths.size();
     for (int i = 0; i < image_paths.size(); i += FLAGS_batch_size) {
-      auto start = system_clock::now();
       int im_vec_size =
           std::min(static_cast<int>(image_paths.size()), i + FLAGS_batch_size);
       std::vector<cv::Mat> im_vec(im_vec_size - i);
@@ -92,21 +87,11 @@ int main(int argc, char** argv) {
       for (int j = i; j < im_vec_size; ++j) {
         im_vec[j - i] = std::move(cv::imread(image_paths[j], 1));
       }
-      auto imread_end = system_clock::now();
       model.predict(im_vec, &results, thread_num);
-      auto imread_duration = duration_cast<microseconds>(imread_end - start);
-      total_imread_time_s += static_cast<double>(imread_duration.count()) *
-                             microseconds::period::num /
-                             microseconds::period::den;
-      auto end = system_clock::now();
-      auto duration = duration_cast<microseconds>(end - start);
-      total_running_time_s += static_cast<double>(duration.count()) *
-                              microseconds::period::num /
-                              microseconds::period::den;
       // 可视化
       for (int j = 0; j < im_vec_size - i; ++j) {
         cv::Mat vis_img =
-            PaddleX::Visualize(im_vec[j], results[j], model.labels, colormap);
+            PaddleX::Visualize(im_vec[j], results[j], model.labels);
         std::string save_path =
             PaddleX::generate_save_path(FLAGS_save_dir, image_paths[i + j]);
         cv::imwrite(save_path, vis_img);
@@ -114,28 +99,16 @@ int main(int argc, char** argv) {
       }
     }
   } else {
-    auto start = system_clock::now();
     PaddleX::SegResult result;
     cv::Mat im = cv::imread(FLAGS_image, 1);
     model.predict(im, &result);
-    auto end = system_clock::now();
-    auto duration = duration_cast<microseconds>(end - start);
-    total_running_time_s += static_cast<double>(duration.count()) *
-                            microseconds::period::num /
-                            microseconds::period::den;
     // 可视化
-    cv::Mat vis_img = PaddleX::Visualize(im, result, model.labels, colormap);
+    cv::Mat vis_img = PaddleX::Visualize(im, result, model.labels);
     std::string save_path =
         PaddleX::generate_save_path(FLAGS_save_dir, FLAGS_image);
     cv::imwrite(save_path, vis_img);
     result.clear();
     std::cout << "Visualized output saved as " << save_path << std::endl;
   }
-  std::cout << "Total running time: " << total_running_time_s
-            << " s, average running time: " << total_running_time_s / imgs
-            << " s/img, total read img time: " << total_imread_time_s
-            << " s, average read img time: " << total_imread_time_s / imgs
-            << " s, batch_size = " << FLAGS_batch_size << std::endl;
-
   return 0;
 }

deploy/cpp/include/paddlex/visualize.h

@@ -65,13 +65,12 @@ std::vector<int> GenerateColorMap(int num_class);
  * @param img: initial image matrix
  * @param results: the detection result
  * @param labels: label map
- * @param colormap: visualization color map
+ * @param threshold: minimum confidence to display
  * @return visualized image matrix
  * */
 cv::Mat Visualize(const cv::Mat& img,
                      const DetResult& results,
                      const std::map<int, std::string>& labels,
-                     const std::vector<int>& colormap,
                      float threshold = 0.5);
 
 /*
@@ -81,13 +80,11 @@ cv::Mat Visualize(const cv::Mat& img,
  * @param img: initial image matrix
  * @param results: the detection result
  * @param labels: label map
- * @param colormap: visualization color map
  * @return visualized image matrix
  * */
 cv::Mat Visualize(const cv::Mat& img,
                      const SegResult& result,
-                     const std::map<int, std::string>& labels,
-                     const std::vector<int>& colormap);
+                     const std::map<int, std::string>& labels);
 
 /*
  * @brief

deploy/cpp/src/visualize.cpp

@@ -34,8 +34,8 @@ std::vector<int> GenerateColorMap(int num_class) {
 cv::Mat Visualize(const cv::Mat& img,
                      const DetResult& result,
                      const std::map<int, std::string>& labels,
-                     const std::vector<int>& colormap,
                      float threshold) {
+  auto colormap = GenerateColorMap(labels.size());
   cv::Mat vis_img = img.clone();
   auto boxes = result.boxes;
   for (int i = 0; i < boxes.size(); ++i) {
@@ -107,8 +107,8 @@ cv::Mat Visualize(const cv::Mat& img,
 
 cv::Mat Visualize(const cv::Mat& img,
                      const SegResult& result,
-                     const std::map<int, std::string>& labels,
-                     const std::vector<int>& colormap) {
+                     const std::map<int, std::string>& labels) {
+  auto colormap = GenerateColorMap(labels.size());
   std::vector<uint8_t> label_map(result.label_map.data.begin(),
                                  result.label_map.data.end());
   cv::Mat mask(result.label_map.shape[0],