Pre Merge pull request !14 from Daniel/lxm

CMakeLists.txt

@@ -275,6 +275,9 @@ target_link_libraries(GimbalUdpDetectionInfoSender sv_world)
 add_executable(EvalFpsOnVideo samples/test/eval_fps_on_video.cpp)
 target_link_libraries(EvalFpsOnVideo sv_world)
 
+add_executable(EvalModelOnCocoVal samples/test/eval_mAP_on_coco_val/eval_mAP_on_coco_val.cpp)
+target_link_libraries(EvalModelOnCocoVal sv_world)
+
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/samples/calib)
 add_executable(CameraCalibrarion samples/calib/calibrate_camera_charuco.cpp)
 target_link_libraries(CameraCalibrarion ${OpenCV_LIBS})

samples/test/eval_mAP_on_coco_val/coco_eval.py

@@ -0,0 +1,25 @@
+from pycocotools.coco import COCO
+from pycocotools.cocoeval import COCOeval
+import os
+
+if __name__ == '__main__':
+    path = os.path.abspath(os.path.join(os.getcwd(),"../../.."))
+    pred_json = 'pd_coco.json'
+    anno_json = path + '/val2017/instances_val2017.json'
+
+    # use COCO API to load forecast results and annotations
+    cocoGt = COCO(anno_json)
+    cocoDt = cocoGt.loadRes(pred_json)
+
+    # create COCO eval object
+    cocoEval = COCOeval(cocoGt, cocoDt,'bbox')
+
+    # assessment
+    cocoEval.evaluate()
+    cocoEval.accumulate()
+    cocoEval.summarize()
+
+
+    # save results
+    with open('coco_eval.txt', 'w') as f:
+        f.write(str(cocoEval.stats))

samples/test/eval_mAP_on_coco_val/eval_mAP_on_coco_val.cpp

@@ -0,0 +1,100 @@
+#include <iostream>
+#include <string>
+// 包含SpireCV SDK头文件
+#include <sv_world.h>
+
+using namespace std;
+using namespace cv;
+
+//extract name
+std::string GetImageFileName(const std::string& imagePath) {
+    size_t lastSlash = imagePath.find_last_of("/\\");
+    if (lastSlash == std::string::npos) {
+        return imagePath;
+    } else {
+        std::string fileName = imagePath.substr(lastSlash + 1);
+        size_t lastDot = fileName.find_last_of(".");
+        if (lastDot != std::string::npos) {
+            return fileName.substr(0, lastDot);
+        }
+        return fileName;
+    }
+}
+
+
+int main(int argc, char *argv[])
+{
+  // 实例化 通用目标 检测器类
+  sv::CommonObjectDetector cod;
+  // 手动导入相机参数，如果使用Amov的G1等吊舱或相机，则可以忽略该步骤，将自动下载相机参数文件
+  cod.loadCameraParams(sv::get_home() + "/SpireCV/calib_webcam_640x480.yaml");
+
+  //load data
+  string val_path = sv::get_home() + "/SpireCV/val2017/val2017"; 
+  vector<string> val_image;
+  glob(val_path, val_image, false);
+  if (val_image.size() == 0)
+  { 
+    printf("val_image error!!!\n");
+    exit(1);
+  }
+
+  //preds folder
+
+  std::string folder = sv::get_home() + "/SpireCV/val2017/preds";
+  int checkStatus = std::system(("if [ -d \"" + folder + "\" ]; then echo; fi").c_str());
+  if(checkStatus == 0)
+  {
+    int removeStatus = std::system(("rm -rf \"" + folder + "\"").c_str());
+    if(removeStatus != 0)
+    {
+      printf("remove older preds folder error!!!\n");
+      exit(1);
+    }
+  } 
+
+  int status = std::system(("mkdir \""+folder+"\"").c_str());
+  if(status != 0)
+  {
+    printf("create preds folder error!!!\n");
+    exit(1);
+  }
+
+
+  for (int i = 0; i < val_image.size(); i++) {
+
+    //create pred file
+    std::string val_image_name = GetImageFileName(val_image[i]);
+    std::string filename = folder+"/"+ val_image_name + ".txt";
+    std::ofstream file(filename);
+    file.is_open();
+    file<<std::fixed<<std::setprecision(6);
+
+    // 实例化SpireCV的 单帧检测结果 接口类 TargetsInFrame
+    sv::TargetsInFrame tgts(i);
+    cv::Mat img = imread(val_image[i]);
+
+    int rows = img.rows;
+    int cols = img.cols;
+   
+    // 执行通用目标检测
+    cod.detect(img, tgts);
+    // 可视化检测结果，叠加到img上
+    sv::drawTargetsInFrame(img, tgts);
+    
+    // reslusts
+    for (int j = 0; j < tgts.targets.size(); j++)
+    {
+        sv::Box b;
+        tgts.targets[j].getBox(b);
+        file<<tgts.targets[j].category_id<<" "<<(float)(b.x1+b.x2)/(2*cols)<<" "<<(float)(b.y1+b.y2)/(2*rows)<<" "<<(float)(b.x2-b.x1)/cols<<" "<<(float)(b.y2-b.y1)/rows<<" "<<(float)tgts.targets[j].score<<"\n";
+    }
+    file.close();
+    
+    cv::imshow("image", img);
+    cv::waitKey(500);
+    img.release();
+    cv::destroyAllWindows();
+    }
+  return 0;
+}

samples/test/eval_mAP_on_coco_val/pd2cocojson.py

@@ -0,0 +1,62 @@
+import datetime
+import json
+import os
+import cv2
+
+# revert prediction results to coco_json 
+path = os.path.abspath(os.path.join(os.getcwd(),"../../.."))
+# all files dir
+images_path = path + '/val2017/val2017'
+preds_path = path + '/val2017/preds'
+
+coco_json_save = 'pd_coco.json'
+
+# config coco_json
+coco_json = []
+
+# remap the id of the coco dataset
+id_map = {1: 0, 2: 1, 3: 2, 4: 3, 5: 4, 6: 5, 7: 6, 8: 7, 9: 8, 10: 9, 11: 10, 13: 11, 14: 12, 15: 13, 16: 14, 17: 15, 18: 16, 19: 17, 20: 18, 21: 19, 22: 20, 23: 21, 24: 22, 25: 23, 27: 24, 28: 25, 31: 26, 32: 27, 33: 28, 34: 29, 35: 30, 36: 31, 37: 32, 38: 33, 39: 34, 40: 35, 41: 36, 42: 37, 43: 38, 44: 39, 46: 40, 47: 41, 48: 42, 49: 43, 50: 44, 51: 45, 52: 46, 53: 47, 54: 48, 55: 49, 56: 50, 57: 51, 58: 52, 59: 53, 60: 54, 61: 55, 62: 56, 63: 57, 64: 58, 65: 59, 67: 60, 70: 61, 72: 62, 73: 63, 74: 64, 75: 65, 76: 66, 77: 67, 78: 68, 79: 69, 80: 70, 81: 71, 82: 72, 84: 73, 85: 74, 86: 75, 87: 76, 88: 77, 89: 78, 90: 79}
+reid_mp = {value: key for key, value in id_map.items()}
+
+# load images dir
+images = os.listdir(images_path)
+for image in images:
+    print(image)
+    # get image name
+    image_name, image_suffix = os.path.splitext(image)
+    # get image W and H
+    image_path = images_path + '/' + image
+    img = cv2.imread(image_path)
+    height, width, _ = img.shape
+
+    # read pred's txt
+    pred_path = preds_path + '/' + image_name + '.txt'
+    if not os.path.exists(pred_path):
+        continue
+    with open(pred_path, 'r') as f:
+        preds = f.readlines()
+        preds = [l.strip() for l in preds]
+        for j, pred in enumerate(preds):
+            pred = pred.split(' ')
+            category_id = int(pred[0])
+            x = float(pred[1]) * width
+            y = float(pred[2]) * height
+            w = float(pred[3]) * width
+            h = float(pred[4]) * height
+            xmin = x - w / 2
+            ymin = y - h / 2
+            xmax = x + w / 2
+            ymax = y + h / 2
+			
+            coco_json.append({
+                'image_id': int(image_name),
+                'category_id': int(reid_mp[category_id]),
+                'bbox': [xmin, ymin, w, h],
+                'score': float(pred[5]),
+                'area': w * h})
+
+# save json
+with open(os.path.join(coco_json_save), 'w') as f:
+    json.dump(coco_json, f, indent=2)
+
+print(len(coco_json), 'Done!')