add visual result style

README.md

@@ -41,6 +41,45 @@ b). If you need full capacity which thor does, including `vis`, `geometry`, `dat
 
 - **2050.01.01**: to be continue..
 
+- **2019.11.16**: We demonstrate how to using `thor::vis` to draw detections in your algorithm:
+
+    ![](https://s2.ax1x.com/2019/11/15/Mdig8e.png)
+
+    ```c++
+    #include "thor/dl.h"
+    #include "thor/os.h"
+    #include "thor/structures.h"
+    #include "thor/vis.h"
+
+    using namespace std;
+    using namespace cv;
+    using namespace thor::vis;
+    using namespace thor::dl;
+
+    int main() {
+        vector<thor::Box> all_detections;
+      for (int i = 0; i < num_det; i++) {
+        // Show results over confidence threshold
+        if (scores[i] >= 0.33f) {
+          float x1 = boxes[i * 4 + 0] * scale_x;
+          float y1 = boxes[i * 4 + 1] * scale_y;
+          float x2 = boxes[i * 4 + 2] * scale_x;
+          float y2 = boxes[i * 4 + 3] * scale_y;
+          thor::Box one_box{x1, y1, x2, y2, thor::BoxFormat::XYXY};
+          one_box.score = scores[i];
+          one_box.idx = classes[i] + 1;
+          all_detections.emplace_back(one_box);
+        }
+      }
+
+    // draw
+    auto res_image = thor::vis::VisualizeDetectionStyleDetectron2(
+              image, all_detections, COCO_CLASSES);
+    }
+    ```
+
+    Above is a simple example using `thor::vis::VisualizeDetectionStyleDetectron2` draw bounding boxes using detectron2 style.  From our experiences, these steps time cost is about **0.0001**s, so it's doesn't matter if you generate your box format to `thor::Box` format first and then call our drawing method.
+
 - **2019.09.24**: I just notice thor has been recommended by 爱可可老师 through weibo, here is link: https://weibo.com/1402400261/I8p1gnIkK , check it out!!! If you like this project, pls give a star!!
 
 - **2019.08.29**: A new header-only args parser lib has been added into thor. Original implementation from [here](https://github.com/Taywee/args). We did some changes than original, roughly usage can be used like this:

include/vis.h

@@ -69,6 +69,11 @@ cv::Mat VisualizeDetection(cv::Mat &img, vector<thor::Box> detections,
                            bool enable_mask = true,
                            float confidence_threshold = 0.02,
                            bool normalized = false);
+cv::Mat VisualizeDetectionStyleDetectron2(cv::Mat &img, vector<thor::Box> detections,
+                           vector<string> classes_names,
+                           bool enable_mask = true,
+                           float confidence_threshold = 0.02,
+                           bool normalized = false);
 
 // adding render HumanPose on image
 void renderHumanPose(std::vector<HumanPose> &poses, cv::Mat &image);

src/vis.cpp

@@ -415,6 +415,58 @@ cv::Mat VisualizeDetection(cv::Mat &img, vector<thor::Box> detections, vector<st
   // maybe combine a mask img back later
   return combined;
 }
+
+cv::Mat VisualizeDetectionStyleDetectron2(cv::Mat &img, vector<thor::Box> detections, vector<string> classes_names, bool enable_mask, float confidence_threshold, bool normalized){
+  // for detectron2 style drawing bounding boxes
+  const int font = cv::FONT_HERSHEY_SIMPLEX;
+  const float font_scale = 0.35;
+  const int font_thickness = 1;
+  cv::Mat mask = cv::Mat::zeros(img.size(), CV_8UC3);
+  for (int i = 0; i < detections.size(); ++i)
+  {
+    thor::Box box = detections[i];
+    box.to_xyxy();
+    const float score = box.score;
+    if (score >= confidence_threshold)
+    {
+      cv::Point pt1, pt2;
+      if (normalized)
+      {
+        pt1.x = (img.cols * box.xmin);
+        pt1.y = (img.rows * box.ymin);
+        pt2.x = (img.cols * box.xmax);
+        pt2.y = (img.rows * box.ymax);
+      }
+      else
+      {
+        pt1.x = box.xmin;
+        pt1.y = box.ymin;
+        pt2.x = box.xmax;
+        pt2.y = box.ymax;
+      }
+
+      cv::Scalar u_c = thor::vis::gen_unique_color_cv(box.idx);
+      cv::rectangle(img, pt1, pt2, u_c, 1, 8, 0);
+      cv::rectangle(mask, pt1, pt2, u_c, cv::FILLED, 0);
+
+      char score_str[256];
+      sprintf(score_str, "%.1f", score);
+      std::string label_text = classes_names[box.idx] + ":" + string(score_str);
+      int base_line = 0;
+      cv::Point text_origin = cv::Point(pt1.x, pt1.y-2);
+      cv::Size text_size = cv::getTextSize(label_text, font, font_scale, font_thickness, &base_line);
+      cv::rectangle(img, cv::Point(text_origin.x, text_origin.y),
+                    cv::Point(text_origin.x + text_size.width, text_origin.y - text_size.height),
+                    cv::Scalar(0, 0, 0), -1, 0);
+      cv::putText(img, label_text, text_origin, font, font_scale, cv::Scalar(255, 255, 255), font_thickness);
+    }
+  }
+  cv::Mat combined;
+  cv::addWeighted(img, 0.8, mask, 0.6, 0.6, combined);
+  // maybe combine a mask img back later
+  return combined;
+};
+
 #endif
 
 } // namespace vis