KITTI数据集评估方法小结

建议先通篇阅读，了解需要修改什么、文件如何组织等，再进行实践，因为涉及到一些代码的问题。你既可以使用.cpp代码+Linux系统的原汁原味操作，也可以选择python代码直接评估的方式，便捷、优雅。

1 平台与资源

Windows上我试过Visual Studio 2022编译、下载安装CMake、cmd操作等，不是报头文件错误，就是Boost缺少，反正过程就不说了，之后我才知道缺少的那些好多包其实是Linux系统上自带的。最后是在腾讯云上购买了Ubuntu系统一年使用权，正好最近春节有活动，hh

链接：(1)github上较多的evaluate_object_3d_offline文件以及(2)我后面主要使用的CSDN博主修改evaluate_object文件，两个都可以。

2 编译与运行

编译的时候在放置.cpp文件与.h文件的目录下运行指令，主要参考链接(2)这位老哥的方法：

cmake .

make

g++ -O3 -DNDEBUG -o evaluate_object evaluate_object.cpp

一步一步来，反正基本上每一步都会报错，报啥错、缺啥包，直接去网上搜ubuntu系统安装某个包的方法 (我也是小白)。

编译好了后没有问题了，会生成一个不带后缀的evaluate_object文件，下面就只剩下数据准备(可以先进行)与运行了。另外，我只是第一次用了cmake .与 make 指令，后面再编译的时候发现只用第三行指令也可以编译好。

3 文件格式及目录结构

在运行之前，我们还需要准备数据噻，label与results。

3.1 格式

labels就是kitti数据集里面的label_2，选择你需要的N个测试文件，如3769个，格式如下：

type	truncated	occluded	alpha	bbox	dimensions	location	rotation_y
Pedestrian	-1	-1	0.29	873.70 152.10 933.44 256.07	1.87 0.50 0.90	5.42 1.50 13.43	0.67

预测结果格式如下：(注意，最后一个数字是预测结果特有的score)

type	truncated	occluded	alpha	bbox	dimensions	location	rotation_y	score
Pedestrian	-1	-1	0.29	873.70 152.10 933.44 256.07	1.87 0.50 0.90	5.42 1.50 13.43	0.67	0.99

The detection format should be simillar to the KITTI dataset label format with 15 columns representing:

Values	Name	Description
1	type	Describes the type of object: ‘Car’, ‘Van’, ‘Truck’, ‘Pedestrian’, ‘Person_sitting’, ‘Cyclist’, ‘Tram’, ‘Misc’ or ‘DontCare’
1	truncated	-1
1	occluded	-1
1	alpha	Observation angle of object, ranging [-pi…pi]
4	bbox	2D bounding box of object in the image (0-based index): contains left, top, right, bottom pixel coordinates
3	dimensions	3D object dimensions: height, width, length (in meters)
3	location	3D object location x,y,z in camera coordinates (in meters)
1	rotation_y	Rotation ry around Y-axis in camera coordinates [-pi…pi]
1	score	Only for results: Float, indicating confidence in detection, needed for p/r curves, higher is better.

3.2 文件结构

evaluate_object
evaluate_object.cpp
mail.h
label
   --000001.txt
   --000002.txt
   -- ...
results
   --data
      --000001.txt
      --000002.txt
      -- ...

4 运行

./evaluate_object label results

成功的话，会在results目录下生成plot文件夹和一些 .txt 文件，我一般只查看 result.txt 来记录。

5 代码

参见博客文章(2)

#include 
#include 
#include 
#include 
#include 
#include 
#include 
#include 

#include 

#include 
#include 

#include 
#include 
#include 
#include 

#include "mail.h"

BOOST_GEOMETRY_REGISTER_C_ARRAY_CS(cs::cartesian)

typedef boost::geometry::model::polygon<boost::geometry::model::d2::point_xy<double> > Polygon;


using namespace std;

/*=======================================================================
STATIC EVALUATION PARAMETERS
=======================================================================*/

// holds the number of test images on the server
const int32_t N_TESTIMAGES = 7480;

// easy, moderate and hard evaluation level
enum DIFFICULTY{EASY=0, MODERATE=1, HARD=2};

// evaluation metrics: image, ground or 3D
enum METRIC{IMAGE=0, GROUND=1, BOX3D=2};

// evaluation parameter
const int32_t MIN_HEIGHT[3]     = {40, 25, 25};     // minimum height for evaluated groundtruth/detections
const int32_t MAX_OCCLUSION[3]  = {0, 1, 2};        // maximum occlusion level of the groundtruth used for evaluation
const double  MAX_TRUNCATION[3] = {0.15, 0.3, 0.5}; // maximum truncation level of the groundtruth used for evaluation

// evaluated object classes
enum CLASSES{CAR=0, PEDESTRIAN=1, CYCLIST=2};
const int NUM_CLASS = 3;

// parameters varying per class
vector<string> CLASS_NAMES;
// the minimum overlap required for 2D evaluation on the image/ground plane and 3D evaluation
const double MIN_OVERLAP[3][3] = {{0.7, 0.5, 0.5}, {0.5, 0.25, 0.25}, {0.5, 0.25, 0.25}};
// const double MIN_OVERLAP[3][3] = {{0.7, 0.5, 0.5}, {0.7, 0.5, 0.5}, {0.7, 0.5, 0.5}};

// no. of recall steps that should be evaluated (discretized)
const double N_SAMPLE_PTS = 41;


// initialize class names
void initGlobals () {
  CLASS_NAMES.push_back("car");
  CLASS_NAMES.push_back("pedestrian");
  CLASS_NAMES.push_back("cyclist");
}

/*=======================================================================
DATA TYPES FOR EVALUATION
=======================================================================*/

// holding data needed for precision-recall and precision-aos
struct tPrData {
  vector<double> v;           // detection score for computing score thresholds
  double         similarity;  // orientation similarity
  int32_t        tp;          // true positives
  int32_t        fp;          // false positives
  int32_t        fn;          // false negatives
  tPrData () :
    similarity(0), tp(0), fp(0), fn(0) {}
};

// holding bounding boxes for ground truth and detections
struct tBox {
  string  type;     // object type as car, pedestrian or cyclist,...
  double   x1;      // left corner
  double   y1;      // top corner
  double   x2;      // right corner
  double   y2;      // bottom corner
  double   alpha;   // image orientation
  tBox (string type, double x1,double y1,double x2,double y2,double alpha) :
    type(type),x1(x1),y1(y1),x2(x2),y2(y2),alpha(alpha) {}
};

// holding ground truth data
struct tGroundtruth {
  tBox    box;        // object type, box, orientation
  double  truncation; // truncation 0..1
  int32_t occlusion;  // occlusion 0,1,2 (non, partly, fully)
  double ry;
  double  t1, t2, t3;
  double h, w, l;
  tGroundtruth () :
    box(tBox("invalild",-1,-1,-1,-1,-10)),truncation(-1),occlusion(-1) {}
  tGroundtruth (tBox box,double truncation,int32_t occlusion) :
    box(box),truncation(truncation),occlusion(occlusion) {}
  tGroundtruth (string type,double x1,double y1,double x2,double y2,double alpha,double truncation,int32_t occlusion) :
    box(tBox(type,x1,y1,x2,y2,alpha)),truncation(truncation),occlusion(occlusion) {}
};

// holding detection data
struct tDetection {
  tBox    box;    // object type, box, orientation
  double  thresh; // detection score
  double  ry;
  double  t1, t2, t3;
  double  h, w, l;
  tDetection ():
    box(tBox("invalid",-1,-1,-1,-1,-10)),thresh(-1000) {}
  tDetection (tBox box,double thresh) :
    box(box),thresh(thresh) {}
  tDetection (string type,double x1,double y1,double x2,double y2,double alpha,double thresh) :
    box(tBox(type,x1,y1,x2,y2,alpha)),thresh(thresh) {}
};


/*=======================================================================
FUNCTIONS TO LOAD DETECTION AND GROUND TRUTH DATA ONCE, SAVE RESULTS
=======================================================================*/
vector<int32_t> indices;

vector<tDetection> loadDetections(string file_name, bool &compute_aos,
        vector<bool> &eval_image, vector<bool> &eval_ground,
        vector<bool> &eval_3d, bool &success) {

  // holds all detections (ignored detections are indicated by an index vector
  vector<tDetection> detections;
  FILE *fp = fopen(file_name.c_str(),"r");
  if (!fp) {
    success = false;
    return detections;
  }
  while (!feof(fp)) {
    tDetection d;
    double trash;
    char str[255];
    if (fscanf(fp, "%s %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf",
                   str, &trash, &trash, &d.box.alpha, &d.box.x1, &d.box.y1,
                   &d.box.x2, &d.box.y2, &d.h, &d.w, &d.l, &d.t1, &d.t2, &d.t3,
                   &d.ry, &d.thresh)==16) {

        // d.thresh = 1;
      d.box.type = str;
      detections.push_back(d);

      // orientation=-10 is invalid, AOS is not evaluated if at least one orientation is invalid
      if(d.box.alpha == -10)
        compute_aos = false;

      // a class is only evaluated if it is detected at least once
      for (int c = 0; c < NUM_CLASS; c++) {
        if (!strcasecmp(d.box.type.c_str(), CLASS_NAMES[c].c_str())) {
          if (!eval_image[c] && d.box.x1 >= 0)
            eval_image[c] = true;
          if (!eval_ground[c] && d.t1 != -1000)
            eval_ground[c] = true;
          if (!eval_3d[c] && d.t2 != -1000)
            eval_3d[c] = true;
          break;
        }
      }
    }
  }
  fclose(fp);
  success = true;
  return detections;
}

vector<tGroundtruth> loadGroundtruth(string file_name,bool &success) {

  // holds all ground truth (ignored ground truth is indicated by an index vector
  vector<tGroundtruth> groundtruth;
  FILE *fp = fopen(file_name.c_str(),"r");
  if (!fp) {
    success = false;
    return groundtruth;
  }
  while (!feof(fp)) {
    tGroundtruth g;
    char str[255];
    if (fscanf(fp, "%s %lf %d %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf",
                   str, &g.truncation, &g.occlusion, &g.box.alpha,
                   &g.box.x1,   &g.box.y1,     &g.box.x2,    &g.box.y2,
                   &g.h,      &g.w,        &g.l,       &g.t1,
                   &g.t2,      &g.t3,        &g.ry )==15) {
      g.box.type = str;
      groundtruth.push_back(g);
    }
  }
  fclose(fp);
  success = true;
  return groundtruth;
}

void saveStats (const vector<double> &precision, const vector<double> &aos, FILE *fp_det, FILE *fp_ori) {

  // save precision to file
  if(precision.empty())
    return;
  for (int32_t i=0; i<precision.size(); i++)
    fprintf(fp_det,"%f ",precision[i]);
  fprintf(fp_det,"\n");

  // save orientation similarity, only if there were no invalid orientation entries in submission (alpha=-10)
  if(aos.empty())
    return;
  for (int32_t i=0; i<aos.size(); i++)
    fprintf(fp_ori,"%f ",aos[i]);
  fprintf(fp_ori,"\n");
}

/*=======================================================================
EVALUATION HELPER FUNCTIONS
=======================================================================*/

// criterion defines whether the overlap is computed with respect to both areas (ground truth and detection)
// or with respect to box a or b (detection and "dontcare" areas)
inline double imageBoxOverlap(tBox a, tBox b, int32_t criterion=-1){

  // overlap is invalid in the beginning
  double o = -1;

  // get overlapping area
  double x1 = max(a.x1, b.x1);
  double y1 = max(a.y1, b.y1);
  double x2 = min(a.x2, b.x2);
  double y2 = min(a.y2, b.y2);

  // compute width and height of overlapping area
  double w = x2-x1;
  double h = y2-y1;

  // set invalid entries to 0 overlap
  if(w<=0 || h<=0)
    return 0;

  // get overlapping areas
  double inter = w*h;
  double a_area = (a.x2-a.x1) * (a.y2-a.y1);
  double b_area = (b.x2-b.x1) * (b.y2-b.y1);

  // intersection over union overlap depending on users choice
  if(criterion==-1)     // union
    o = inter / (a_area+b_area-inter);
  else if(criterion==0) // bbox_a
    o = inter / a_area;
  else if(criterion==1) // bbox_b
    o = inter / b_area;

  // overlap
  return o;
}

inline double imageBoxOverlap(tDetection a, tGroundtruth b, int32_t criterion=-1){
  return imageBoxOverlap(a.box, b.box, criterion);
}

// compute polygon of an oriented bounding box
template <typename T>
Polygon toPolygon(const T& g) {
    using namespace boost::numeric::ublas;
    using namespace boost::geometry;
    matrix<double> mref(2, 2);
    mref(0, 0) = cos(g.ry); mref(0, 1) = sin(g.ry);
    mref(1, 0) = -sin(g.ry); mref(1, 1) = cos(g.ry);

    static int count = 0;
    matrix<double> corners(2, 4);
    double data[] = {g.l / 2, g.l / 2, -g.l / 2, -g.l / 2,
                     g.w / 2, -g.w / 2, -g.w / 2, g.w / 2};
    std::copy(data, data + 8, corners.data().begin());
    matrix<double> gc = prod(mref, corners);
    for (int i = 0; i < 4; ++i) {
        gc(0, i) += g.t1;
        gc(1, i) += g.t3;
    }

    double points[][2] = {{gc(0, 0), gc(1, 0)},{gc(0, 1), gc(1, 1)},{gc(0, 2), gc(1, 2)},{gc(0, 3), gc(1, 3)},{gc(0, 0), gc(1, 0)}};
    Polygon poly;
    append(poly, points);
    return poly;
}

// measure overlap between bird's eye view bounding boxes, parametrized by (ry, l, w, tx, tz)
inline double groundBoxOverlap(tDetection d, tGroundtruth g, int32_t criterion = -1) {
    using namespace boost::geometry;
    Polygon gp = toPolygon(g);
    Polygon dp = toPolygon(d);

    std::vector<Polygon> in, un;
    intersection(gp, dp, in);
    union_(gp, dp, un);

    double inter_area = in.empty() ? 0 : area(in.front());
    double union_area = area(un.front());
    double o;
    if(criterion==-1)     // union
        o = inter_area / union_area;
    else if(criterion==0) // bbox_a
        o = inter_area / area(dp);
    else if(criterion==1) // bbox_b
        o = inter_area / area(gp);

    return o;
}

// measure overlap between 3D bounding boxes, parametrized by (ry, h, w, l, tx, ty, tz)
inline double box3DOverlap(tDetection d, tGroundtruth g, int32_t criterion = -1) {
    using namespace boost::geometry;
    Polygon gp = toPolygon(g);
    Polygon dp = toPolygon(d);

    std::vector<Polygon> in, un;
    intersection(gp, dp, in);
    union_(gp, dp, un);

    double ymax = min(d.t2, g.t2);
    double ymin = max(d.t2 - d.h, g.t2 - g.h);

    double inter_area = in.empty() ? 0 : area(in.front());
    double inter_vol = inter_area * max(0.0, ymax - ymin);

    double det_vol = d.h * d.l * d.w;
    double gt_vol = g.h * g.l * g.w;

    double o;
    if(criterion==-1)     // union
        o = inter_vol / (det_vol + gt_vol - inter_vol);
    else if(criterion==0) // bbox_a
        o = inter_vol / det_vol;
    else if(criterion==1) // bbox_b
        o = inter_vol / gt_vol;

    return o;
}

vector<double> getThresholds(vector<double> &v, double n_groundtruth){

  // holds scores needed to compute N_SAMPLE_PTS recall values
  vector<double> t;

  // sort scores in descending order
  // (highest score is assumed to give best/most confident detections)
  sort(v.begin(), v.end(), greater<double>());

  // get scores for linearly spaced recall
  double current_recall = 0;
  for(int32_t i=0; i<v.size(); i++){

    // check if right-hand-side recall with respect to current recall is close than left-hand-side one
    // in this case, skip the current detection score
    double l_recall, r_recall, recall;
    l_recall = (double)(i+1)/n_groundtruth;
    if(i<(v.size()-1))
      r_recall = (double)(i+2)/n_groundtruth;
    else
      r_recall = l_recall;

    if( (r_recall-current_recall) < (current_recall-l_recall) && i<(v.size()-1))
      continue;

    // left recall is the best approximation, so use this and goto next recall step for approximation
    recall = l_recall;

    // the next recall step was reached
    t.push_back(v[i]);
    current_recall += 1.0/(N_SAMPLE_PTS-1.0);
  }
  return t;
}

void cleanData(CLASSES current_class, const vector<tGroundtruth> &gt, const vector<tDetection> &det, vector<int32_t> &ignored_gt, vector<tGroundtruth> &dc, vector<int32_t> &ignored_det, int32_t &n_gt, DIFFICULTY difficulty){

  // extract ground truth bounding boxes for current evaluation class
  for(int32_t i=0;i<gt.size(); i++){

    // only bounding boxes with a minimum height are used for evaluation
    double height = gt[i].box.y2 - gt[i].box.y1;

    // neighboring classes are ignored ("van" for "car" and "person_sitting" for "pedestrian")
    // (lower/upper cases are ignored)
    int32_t valid_class;

    // all classes without a neighboring class
    if(!strcasecmp(gt[i].box.type.c_str(), CLASS_NAMES[current_class].c_str()))
      valid_class = 1;

    // classes with a neighboring class
    else if(!strcasecmp(CLASS_NAMES[current_class].c_str(), "Pedestrian") && !strcasecmp("Person_sitting", gt[i].box.type.c_str()))
      valid_class = 0;
    else if(!strcasecmp(CLASS_NAMES[current_class].c_str(), "Car") && !strcasecmp("Van", gt[i].box.type.c_str()))
      valid_class = 0;

    // classes not used for evaluation
    else
      valid_class = -1;

    // ground truth is ignored, if occlusion, truncation exceeds the difficulty or ground truth is too small
    // (doesn't count as FN nor TP, although detections may be assigned)
    bool ignore = false;
    if(gt[i].occlusion>MAX_OCCLUSION[difficulty] || gt[i].truncation>MAX_TRUNCATION[difficulty] || height<MIN_HEIGHT[difficulty])
      ignore = true;

    // set ignored vector for ground truth
    // current class and not ignored (total no. of ground truth is detected for recall denominator)
    if(valid_class==1 && !ignore){
      ignored_gt.push_back(0);
      n_gt++;
    }

    // neighboring class, or current class but ignored
    else if(valid_class==0 || (ignore && valid_class==1))
      ignored_gt.push_back(1);

    // all other classes which are FN in the evaluation
    else
      ignored_gt.push_back(-1);
  }

  // extract dontcare areas
  for(int32_t i=0;i<gt.size(); i++)
    if(!strcasecmp("DontCare", gt[i].box.type.c_str()))
      dc.push_back(gt[i]);

  // extract detections bounding boxes of the current class
  for(int32_t i=0;i<det.size(); i++){

    // neighboring classes are not evaluated
    int32_t valid_class;
    if(!strcasecmp(det[i].box.type.c_str(), CLASS_NAMES[current_class].c_str()))
      valid_class = 1;
    else
      valid_class = -1;

    int32_t height = fabs(det[i].box.y1 - det[i].box.y2);

    // set ignored vector for detections
    if(height<MIN_HEIGHT[difficulty])
      ignored_det.push_back(1);
    else if(valid_class==1)
      ignored_det.push_back(0);
    else
      ignored_det.push_back(-1);
  }
}

tPrData computeStatistics(CLASSES current_class, const vector<tGroundtruth> &gt,
        const vector<tDetection> &det, const vector<tGroundtruth> &dc,
        const vector<int32_t> &ignored_gt, const vector<int32_t>  &ignored_det,
        bool compute_fp, double (*boxoverlap)(tDetection, tGroundtruth, int32_t),
        METRIC metric, bool compute_aos=false, double thresh=0, bool debug=false){

  tPrData stat = tPrData();
  const double NO_DETECTION = -10000000;
  vector<double> delta;            // holds angular difference for TPs (needed for AOS evaluation)
  vector<bool> assigned_detection; // holds wether a detection was assigned to a valid or ignored ground truth
  assigned_detection.assign(det.size(), false);
  vector<bool> ignored_threshold;
  ignored_threshold.assign(det.size(), false); // holds detections with a threshold lower than thresh if FP are computed

  // detections with a low score are ignored for computing precision (needs FP)
  if(compute_fp)
    for(int32_t i=0; i<det.size(); i++)
      if(det[i].thresh<thresh)
        ignored_threshold[i] = true;

  // evaluate all ground truth boxes
  for(int32_t i=0; i<gt.size(); i++){

    // this ground truth is not of the current or a neighboring class and therefore ignored
    if(ignored_gt[i]==-1)
      continue;

    /*=======================================================================
    find candidates (overlap with ground truth > 0.5) (logical len(det))
    =======================================================================*/
    int32_t det_idx          = -1;
    double valid_detection = NO_DETECTION;
    double max_overlap     = 0;

    // search for a possible detection
    bool assigned_ignored_det = false;
    for(int32_t j=0; j<det.size(); j++){

      // detections not of the current class, already assigned or with a low threshold are ignored
      if(ignored_det[j]==-1)
        continue;
      if(assigned_detection[j])
        continue;
      if(ignored_threshold[j])
        continue;

      // find the maximum score for the candidates and get idx of respective detection
      double overlap = boxoverlap(det[j], gt[i], -1);

      // for computing recall thresholds, the candidate with highest score is considered
      if(!compute_fp && overlap>MIN_OVERLAP[metric][current_class] && det[j].thresh>valid_detection){
        det_idx         = j;
        valid_detection = det[j].thresh;
      }

      // for computing pr curve values, the candidate with the greatest overlap is considered
      // if the greatest overlap is an ignored detection (min_height), the overlapping detection is used
      else if(compute_fp && overlap>MIN_OVERLAP[metric][current_class] && (overlap>max_overlap || assigned_ignored_det) && ignored_det[j]==0){
        max_overlap     = overlap;
        det_idx         = j;
        valid_detection = 1;
        assigned_ignored_det = false;
      }
      else if(compute_fp && overlap>MIN_OVERLAP[metric][current_class] && valid_detection==NO_DETECTION && ignored_det[j]==1){
        det_idx              = j;
        valid_detection      = 1;
        assigned_ignored_det = true;
      }
    }

    /*=======================================================================
    compute TP, FP and FN
    =======================================================================*/

    // nothing was assigned to this valid ground truth
    if(valid_detection==NO_DETECTION && ignored_gt[i]==0) {
      stat.fn++;
    }

    // only evaluate valid ground truth <=> detection assignments (considering difficulty level)
    else if(valid_detection!=NO_DETECTION && (ignored_gt[i]==1 || ignored_det[det_idx]==1))
      assigned_detection[det_idx] = true;

    // found a valid true positive
    else if(valid_detection!=NO_DETECTION){

      // write highest score to threshold vector
      stat.tp++;
      stat.v.push_back(det[det_idx].thresh);

      // compute angular difference of detection and ground truth if valid detection orientation was provided
      if(compute_aos)
        delta.push_back(gt[i].box.alpha - det[det_idx].box.alpha);

      // clean up
      assigned_detection[det_idx] = true;
    }
  }

  // if FP are requested, consider stuff area
  if(compute_fp){

    // count fp
    for(int32_t i=0; i<det.size(); i++){

      // count false positives if required (height smaller than required is ignored (ignored_det==1)
      if(!(assigned_detection[i] || ignored_det[i]==-1 || ignored_det[i]==1 || ignored_threshold[i]))
        stat.fp++;
    }

    // do not consider detections overlapping with stuff area
    int32_t nstuff = 0;
    for(int32_t i=0; i<dc.size(); i++){
      for(int32_t j=0; j<det.size(); j++){

        // detections not of the current class, already assigned, with a low threshold or a low minimum height are ignored
        if(assigned_detection[j])
          continue;
        if(ignored_det[j]==-1 || ignored_det[j]==1)
          continue;
        if(ignored_threshold[j])
          continue;

        // compute overlap and assign to stuff area, if overlap exceeds class specific value
        double overlap = boxoverlap(det[j], dc[i], 0);
        if(overlap>MIN_OVERLAP[metric][current_class]){
          assigned_detection[j] = true;
          nstuff++;
        }
      }
    }

    // FP = no. of all not to ground truth assigned detections - detections assigned to stuff areas
    stat.fp -= nstuff;

    // if all orientation values are valid, the AOS is computed
    if(compute_aos){
      vector<double> tmp;

      // FP have a similarity of 0, for all TP compute AOS
      tmp.assign(stat.fp, 0);
      for(int32_t i=0; i<delta.size(); i++)
        tmp.push_back((1.0+cos(delta[i]))/2.0);

      // be sure, that all orientation deltas are computed
      assert(tmp.size()==stat.fp+stat.tp);
      assert(delta.size()==stat.tp);

      // get the mean orientation similarity for this image
      if(stat.tp>0 || stat.fp>0)
        stat.similarity = accumulate(tmp.begin(), tmp.end(), 0.0);

      // there was neither a FP nor a TP, so the similarity is ignored in the evaluation
      else
        stat.similarity = -1;
    }
  }
  return stat;
}

/*=======================================================================
EVALUATE CLASS-WISE
=======================================================================*/

bool eval_class (FILE *fp_det, FILE *fp_ori, CLASSES current_class,
        const vector< vector<tGroundtruth> > &groundtruth,
        const vector< vector<tDetection> > &detections, bool compute_aos,
        double (*boxoverlap)(tDetection, tGroundtruth, int32_t),
        vector<double> &precision, vector<double> &aos,
        DIFFICULTY difficulty, METRIC metric) {
    assert(groundtruth.size() == detections.size());

  // init
  int32_t n_gt=0;                                     // total no. of gt (denominator of recall)
  vector<double> v, thresholds;                       // detection scores, evaluated for recall discretization
  vector< vector<int32_t> > ignored_gt, ignored_det;  // index of ignored gt detection for current class/difficulty
  vector< vector<tGroundtruth> > dontcare;            // index of dontcare areas, included in ground truth

  // for all test images do
  for (int32_t i=0; i<groundtruth.size(); i++){

    // holds ignored ground truth, ignored detections and dontcare areas for current frame
    vector<int32_t> i_gt, i_det;
    vector<tGroundtruth> dc;

    // only evaluate objects of current class and ignore occluded, truncated objects
    cleanData(current_class, groundtruth[i], detections[i], i_gt, dc, i_det, n_gt, difficulty);
    ignored_gt.push_back(i_gt);
    ignored_det.push_back(i_det);
    dontcare.push_back(dc);

    // compute statistics to get recall values
    tPrData pr_tmp = tPrData();
    pr_tmp = computeStatistics(current_class, groundtruth[i], detections[i], dc, i_gt, i_det, false, boxoverlap, metric);

    // add detection scores to vector over all images
    for(int32_t j=0; j<pr_tmp.v.size(); j++)
      v.push_back(pr_tmp.v[j]);
  }

  // get scores that must be evaluated for recall discretization
  thresholds = getThresholds(v, n_gt);

  // compute TP,FP,FN for relevant scores
  vector<tPrData> pr;
  pr.assign(thresholds.size(),tPrData());
  for (int32_t i=0; i<groundtruth.size(); i++){

    // for all scores/recall thresholds do:
    for(int32_t t=0; t<thresholds.size(); t++){
      tPrData tmp = tPrData();
      tmp = computeStatistics(current_class, groundtruth[i], detections[i], dontcare[i],
                              ignored_gt[i], ignored_det[i], true, boxoverlap, metric,
                              compute_aos, thresholds[t], t==38);

      // add no. of TP, FP, FN, AOS for current frame to total evaluation for current threshold
      pr[t].tp += tmp.tp;
      pr[t].fp += tmp.fp;
      pr[t].fn += tmp.fn;
      if(tmp.similarity!=-1)
        pr[t].similarity += tmp.similarity;
    }
  }

  // compute recall, precision and AOS
  vector<double> recall;
  precision.assign(N_SAMPLE_PTS, 0);
  if(compute_aos)
    aos.assign(N_SAMPLE_PTS, 0);
  double r=0;
  for (int32_t i=0; i<thresholds.size(); i++){
    r = pr[i].tp/(double)(pr[i].tp + pr[i].fn);
    recall.push_back(r);
    precision[i] = pr[i].tp/(double)(pr[i].tp + pr[i].fp);
    if(compute_aos)
      aos[i] = pr[i].similarity/(double)(pr[i].tp + pr[i].fp);
  }

  // filter precision and AOS using max_{i..end}(precision)
  for (int32_t i=0; i<thresholds.size(); i++){
    precision[i] = *max_element(precision.begin()+i, precision.end());
    if(compute_aos)
      aos[i] = *max_element(aos.begin()+i, aos.end());
  }

  // save statisics and finish with success
  saveStats(precision, aos, fp_det, fp_ori);
    return true;
}

float* saveAndPlotPlots(string dir_name,string file_name,string obj_type,vector<double> vals[],bool is_aos){

  char command[1024];

  // save plot data to file
  FILE *fp = fopen((dir_name + "/" + file_name + ".txt").c_str(),"w");
  printf("save %s\n", (dir_name + "/" + file_name + ".txt").c_str());
  for (int32_t i=0; i<(int)N_SAMPLE_PTS; i++)
    fprintf(fp,"%f %f %f %f\n",(double)i/(N_SAMPLE_PTS-1.0),vals[0][i],vals[1][i],vals[2][i]);
  fclose(fp);

  float sum[3] = {0, 0, 0};
  static float results[3] = {0, 0, 0};
  for (int v = 0; v < 3; v++)
      for (int i = 1; i < vals[v].size(); i++)  // 1, 2, 3, ..., 40, 40 points in total
          sum[v] += vals[v][i];
 /*
  if (N_SAMPLE_PTS==11) {
      for (int v = 0; v < 3; ++v)
		  for (int i = 0; i < vals[v].size(); i = i + 4)
			  sum[v] += vals[v][i];
    } else {
      for (int v = 0; v < 3; ++v)
		  for (int i = 1; i < vals[v].size(); i = i + 1)
			  sum[v] += vals[v][i];
    }
*/
  for (int i = 0; i < 3; i++)
      results[i] = sum[i] / (N_SAMPLE_PTS-1.0) * 100;
  printf("%s AP: %f %f %f\n", file_name.c_str(), sum[0] / (N_SAMPLE_PTS-1.0) * 100, sum[1] / (N_SAMPLE_PTS-1.0) * 100, sum[2] / (N_SAMPLE_PTS-1.0) * 100);


  // create png + eps
  for (int32_t j=0; j<2; j++) {

    // open file
    FILE *fp = fopen((dir_name + "/" + file_name + ".gp").c_str(),"w");

    // save gnuplot instructions
    if (j==0) {
      fprintf(fp,"set term png size 450,315 font \"Helvetica\" 11\n");
      fprintf(fp,"set output \"%s.png\"\n",file_name.c_str());
    } else {
      fprintf(fp,"set term postscript eps enhanced color font \"Helvetica\" 20\n");
      fprintf(fp,"set output \"%s.eps\"\n",file_name.c_str());
    }

    // set labels and ranges
    fprintf(fp,"set size ratio 0.7\n");
    fprintf(fp,"set xrange [0:1]\n");
    fprintf(fp,"set yrange [0:1]\n");
    fprintf(fp,"set xlabel \"Recall\"\n");
    if (!is_aos) fprintf(fp,"set ylabel \"Precision\"\n");
    else         fprintf(fp,"set ylabel \"Orientation Similarity\"\n");
    obj_type[0] = toupper(obj_type[0]);
    fprintf(fp,"set title \"%s\"\n",obj_type.c_str());

    // line width
    int32_t   lw = 5;
    if (j==0) lw = 3;

    // plot error curve
    fprintf(fp,"plot ");
    fprintf(fp,"\"%s.txt\" using 1:2 title 'Easy' with lines ls 1 lw %d,",file_name.c_str(),lw);
    fprintf(fp,"\"%s.txt\" using 1:3 title 'Moderate' with lines ls 2 lw %d,",file_name.c_str(),lw);
    fprintf(fp,"\"%s.txt\" using 1:4 title 'Hard' with lines ls 3 lw %d",file_name.c_str(),lw);

    // close file
    fclose(fp);

    // run gnuplot => create png + eps
    sprintf(command,"cd %s; gnuplot %s",dir_name.c_str(),(file_name + ".gp").c_str());
    system(command);
  }

  // create pdf and crop
  sprintf(command,"cd %s; ps2pdf %s.eps %s_large.pdf",dir_name.c_str(),file_name.c_str(),file_name.c_str());
  system(command);
  sprintf(command,"cd %s; pdfcrop %s_large.pdf %s.pdf",dir_name.c_str(),file_name.c_str(),file_name.c_str());
  system(command);
  sprintf(command,"cd %s; rm %s_large.pdf",dir_name.c_str(),file_name.c_str());
  system(command);

  return results;
}

vector<int32_t> getEvalIndices(const string& result_dir) {

    DIR* dir;
    dirent* entity;
    dir = opendir(result_dir.c_str());
    if (dir) {
        while (entity = readdir(dir)) {
            string path(entity->d_name);
            int32_t len = path.size();
            if (len < 10) continue;
            int32_t index = atoi(path.substr(len - 10, 10).c_str());
            indices.push_back(index);
        }
    }
    return indices;
}

bool eval(string gt_dir, string result_dir, Mail* mail){

  // set some global parameters
  initGlobals();

  // ground truth and result directories
  // string gt_dir         = "data/object/label_2";
  // string result_dir     = "results/" + result_sha;
  string plot_dir       = result_dir + "/plot";

  // create output directories
  system(("mkdir " + plot_dir).c_str());

  // hold detections and ground truth in memory
  vector< vector<tGroundtruth> > groundtruth;
  vector< vector<tDetection> >   detections;

  // holds wether orientation similarity shall be computed (might be set to false while loading detections)
  // and which labels where provided by this submission
  bool compute_aos=true;
  vector<bool> eval_image(NUM_CLASS, false);
  vector<bool> eval_ground(NUM_CLASS, false);
  vector<bool> eval_3d(NUM_CLASS, false);

  // for all images read groundtruth and detections
  mail->msg("Loading detections...");
  std::vector<int32_t> indices = getEvalIndices(result_dir + "/data/");
  printf("number of files for evaluation: %d\n", (int)indices.size());

  for (int32_t i=0; i<indices.size(); i++) {

    // file name
    char file_name[256];
    sprintf(file_name,"%06d.txt",indices.at(i));

    // read ground truth and result poses
    bool gt_success,det_success;
    vector<tGroundtruth> gt   = loadGroundtruth(gt_dir + "/" + file_name,gt_success);
    vector<tDetection>   det  = loadDetections(result_dir + "/data/" + file_name,
            compute_aos, eval_image, eval_ground, eval_3d, det_success);
    groundtruth.push_back(gt);
    detections.push_back(det);

    // check for errors
    if (!gt_success) {
      mail->msg("ERROR: Couldn't read: %s of ground truth. Please write me an email!", file_name);
      return false;
    }
    if (!det_success) {
      mail->msg("ERROR: Couldn't read: %s", file_name);
      return false;
    }
  }
  mail->msg("  done.");

  // holds pointers for result files
  FILE *fp_det=0, *fp_ori=0;

  FILE *fp = fopen((result_dir + "/result.txt").c_str(),"w");
  float * results;

  // eval image 2D bounding boxes
  fprintf(fp,"eval image 2D bounding boxes\n");
  for (int c = 0; c < NUM_CLASS; c++) {
    CLASSES cls = (CLASSES)c;
    if (eval_image[c]) {
      fp_det = fopen((result_dir + "/stats_" + CLASS_NAMES[c] + "_detection.txt").c_str(), "w");
      if(compute_aos)
        fp_ori = fopen((result_dir + "/stats_" + CLASS_NAMES[c] + "_orientation.txt").c_str(),"w");
      vector<double> precision[3], aos[3];
      if(   !eval_class(fp_det, fp_ori, cls, groundtruth, detections, compute_aos, imageBoxOverlap, precision[0], aos[0], EASY, IMAGE)
         || !eval_class(fp_det, fp_ori, cls, groundtruth, detections, compute_aos, imageBoxOverlap, precision[1], aos[1], MODERATE, IMAGE)
         || !eval_class(fp_det, fp_ori, cls, groundtruth, detections, compute_aos, imageBoxOverlap, precision[2], aos[2], HARD, IMAGE)) {
        mail->msg("%s evaluation failed.", CLASS_NAMES[c].c_str());
        return false;
      }
      fclose(fp_det);
      results = saveAndPlotPlots(plot_dir, CLASS_NAMES[c] + "_detection", CLASS_NAMES[c], precision, 0);
      fprintf(fp,"%s AP: %f %f %f\n",CLASS_NAMES[c].c_str(), results[0], results[1], results[2]);

      if(compute_aos){
        results = saveAndPlotPlots(plot_dir, CLASS_NAMES[c] + "_orientation", CLASS_NAMES[c], aos, 1);
        fprintf(fp,"%s AP: %f %f %f\n",CLASS_NAMES[c].c_str(), results[0], results[1], results[2]);
        fclose(fp_ori);
      }
    }
  }

  // don't evaluate AOS for birdview boxes and 3D boxes
  compute_aos = false;

  // eval bird's eye view bounding boxes
  fprintf(fp,"\neval bird's eye view bounding boxes\n");
  for (int c = 0; c < NUM_CLASS; c++) {
    CLASSES cls = (CLASSES)c;
    if (eval_ground[c]) {
      fp_det = fopen((result_dir + "/stats_" + CLASS_NAMES[c] + "_detection_ground.txt").c_str(), "w");
      vector<double> precision[3], aos[3];
      if(   !eval_class(fp_det, fp_ori, cls, groundtruth, detections, compute_aos, groundBoxOverlap, precision[0], aos[0], EASY, GROUND)
         || !eval_class(fp_det, fp_ori, cls, groundtruth, detections, compute_aos, groundBoxOverlap, precision[1], aos[1], MODERATE, GROUND)
         || !eval_class(fp_det, fp_ori, cls, groundtruth, detections, compute_aos, groundBoxOverlap, precision[2], aos[2], HARD, GROUND)) {
        mail->msg("%s evaluation failed.", CLASS_NAMES[c].c_str());
        return false;
      }
      fclose(fp_det);
      results = saveAndPlotPlots(plot_dir, CLASS_NAMES[c] + "_detection_ground", CLASS_NAMES[c], precision, 0);
      fprintf(fp,"%s AP: %f %f %f\n",CLASS_NAMES[c].c_str(), results[0], results[1], results[2]);
    }
  }

  // eval 3D bounding boxes
  fprintf(fp,"\neval 3D bounding boxes\n");
  for (int c = 0; c < NUM_CLASS; c++) {
    CLASSES cls = (CLASSES)c;
    if (eval_3d[c]) {
      fp_det = fopen((result_dir + "/stats_" + CLASS_NAMES[c] + "_detection_3d.txt").c_str(), "w");
      vector<double> precision[3], aos[3];
      if(   !eval_class(fp_det, fp_ori, cls, groundtruth, detections, compute_aos, box3DOverlap, precision[0], aos[0], EASY, BOX3D)
         || !eval_class(fp_det, fp_ori, cls, groundtruth, detections, compute_aos, box3DOverlap, precision[1], aos[1], MODERATE, BOX3D)
         || !eval_class(fp_det, fp_ori, cls, groundtruth, detections, compute_aos, box3DOverlap, precision[2], aos[2], HARD, BOX3D)) {
        mail->msg("%s evaluation failed.", CLASS_NAMES[c].c_str());
        return false;
      }
      fclose(fp_det);
      results = saveAndPlotPlots(plot_dir, CLASS_NAMES[c] + "_detection_3d", CLASS_NAMES[c], precision, 0);
      fprintf(fp,"%s AP: %f %f %f\n",CLASS_NAMES[c].c_str(), results[0], results[1], results[2]);
    }
  }

  fclose(fp);

  printf("\n##################################################\n");
  FILE *fp_print = fopen((result_dir + "/result.txt").c_str(),"r");
  char c;
  while(fscanf(fp_print,"%c",&c)!=EOF)
    printf("%c",c);
  fclose(fp_print);
  printf("\n##################################################\n");
  // success
  return true;
}

int32_t main (int32_t argc,char *argv[]) {

  // we need 2 or 4 arguments!
  if (argc!=3) {
    cout << "Usage: ./eval_detection_3d_offline gt_dir result_dir" << endl;
    return 1;
  }

  // read arguments
  string gt_dir = argv[1];
  string result_dir = argv[2];

  // init notification mail
  Mail *mail;
  mail = new Mail();
  mail->msg("Thank you for participating in our evaluation!");

  // run evaluation
  if (eval(gt_dir, result_dir, mail)) {
    mail->msg("Your evaluation results are available at:");
    mail->msg(result_dir.c_str());
  } else {
    system(("rm -r " + result_dir + "/plot").c_str());
    mail->msg("An error occured while processing your results.");
  }

  // send mail and exit
  delete mail;

  return 0;
}

6 注意

1) AP_R11与AP_R40的选择

不知道大家有没有注意到KITTI官网下面有这么一句话：

Note 2: On 08.10.2019, we have followed the suggestions of the Mapillary team in their paper Disentangling Monocular 3D Object Detection and use 40 recall positions instead of the 11 recall positions proposed in the original Pascal VOC benchmark. This results in a more fair comparison of the results, please check their paper. The last leaderboards right before this change can be found here: Object Detection Evaluation, 3D Object Detection Evaluation, Bird’s Eye View Evaluation.

在上面的代码部分，saveAndPlotPlots函数里面有这么一段：

float sum[3] = {0, 0, 0};
  static float results[3] = {0, 0, 0};
  for (int v = 0; v < 3; v++)
      for (int i = 1; i < vals[v].size(); i++)  // 1, 2, 3, ..., 40, 40 points in total
          sum[v] += vals[v][i];
 /*
  if (N_SAMPLE_PTS==11) {
      for (int v = 0; v < 3; ++v)
		  for (int i = 0; i < vals[v].size(); i = i + 4)
			  sum[v] += vals[v][i];
    } else {
      for (int v = 0; v < 3; ++v)
		  for (int i = 1; i < vals[v].size(); i = i + 1)
			  sum[v] += vals[v][i];
    }
*/
  for (int i = 0; i < 3; i++)
      results[i] = sum[i] / (N_SAMPLE_PTS-1.0) * 100;
  printf("%s AP: %f %f %f\n", file_name.c_str(), sum[0] / (N_SAMPLE_PTS-1.0) * 100, sum[1] / (N_SAMPLE_PTS-1.0) * 100, sum[2] / (N_SAMPLE_PTS-1.0) * 100);

而这是用于40个点的插值得到的mAP代码，如果想得11点插值采样的结果，可以改为下面的代码：(只改这里，其他地方不用改)

float sum[3] = {0, 0, 0};
  static float results[3] = {0, 0, 0};
  for (int v = 0; v < 3; v++)
      for (int i = 0; i < vals[v].size(); i=i+4)  // 0, 4, 8, ..., 40, 11 points in total
          sum[v] += vals[v][i];
  for (int i = 0; i < 3; i++)
      results[i] = sum[i] / 11 * 100;  // 注意是11
  printf("%s AP: %f %f %f\n", file_name.c_str(), sum[0] / 11 * 100, sum[1] / 11 * 100, sum[2] / 11 * 100);

亲自动手一遍，相信你也知道它大致是怎么计算的了。

2) MIN_OVERLAP的选择

在evaluate_object.cpp文件里有这么一行代码。

// the minimum overlap required for 2D evaluation on the image/ground plane and 3D evaluation
const double MIN_OVERLAP[3][3] = {{0.7, 0.5, 0.5}, {0.5, 0.25, 0.25}, {0.5, 0.25, 0.25}};
// const double MIN_OVERLAP[3][3] = {{0.7, 0.5, 0.5}, {0.7, 0.5, 0.5}, {0.7, 0.5, 0.5}};

使用第一行IOU threshold的结果比第二行要好一些，一些paper也是用的第一行但没有告诉读者，我一开始用的默认第二行，导致我怀疑自己的评估文件，emmm

========== 后期补充：

7 救苦救难的Python验证代码

只想说 靠(一个动作)！牛！代码我放在我的GitHub与gitee上面，内网外网均可到达。地址：

 1. https://github.com/fyancy/kitti_evaluation_codes
 2. https://gitee.com/fyancy/kitti_evaluation_codes

下面贴出c++代码与OpenPCDet的python代码的比较结果(有误差，但问题不大)：

C++	iou=0.5	40点插值采样

OpenPCDet	iou=0.5	40点插值采样

****Car [email protected], 0.50, 0.50:
bbox AP:90.5915, 78.5379, 69.4228
bev  AP:26.3483, 21.1817, 18.4874
3d   AP:13.1477, 11.7556, 9.7749
aos  AP:86.79, 74.75, 65.51
****Pedestrian [email protected], 0.25, 0.25:
bbox AP:61.7003, 50.6382, 43.5130
bev  AP:15.4725, 12.9547, 11.2692
3d   AP:14.6865, 12.1036, 9.8097
aos  AP:46.16, 37.53, 32.01
****Cyclist [email protected], 0.25, 0.25:
bbox AP:40.1334, 24.1116, 22.8021
bev  AP:9.2079, 4.3063, 4.2338
3d   AP:8.5029, 4.0998, 3.6177
aos  AP:27.46, 16.00, 15.30

C++	iou=0.7	40点插值采样

OpenPCDet	iou=0.7	40点插值采样

****Car [email protected], 0.70, 0.70:
bbox AP:90.5915, 78.5379, 69.4228
bev  AP:1.8337, 1.7438, 1.5608
3d   AP:0.3130, 0.3099, 0.2322
aos  AP:86.79, 74.75, 65.51
****Pedestrian [email protected], 0.50, 0.50:
bbox AP:61.7003, 50.6382, 43.5130
bev  AP:4.3485, 3.3358, 2.7455
3d   AP:3.7477, 2.6003, 1.9591
aos  AP:46.16, 37.53, 32.01
****Cyclist [email protected], 0.50, 0.50:
bbox AP:40.1334, 24.1116, 22.8021
bev  AP:1.1627, 0.5036, 0.3376
3d   AP:0.4191, 0.1398, 0.0595
aos  AP:27.46, 16.00, 15.30

C++	iou=0.5	11点插值采样

OpenPCDet	iou=0.5	11点插值采样

####Car [email protected], 0.50, 0.50:
bbox AP:86.1249, 76.4279, 68.0968
bev  AP:32.1401, 26.3529, 22.4785
3d   AP:19.2953, 18.1446, 15.6865
aos  AP:82.88, 73.04, 64.68
####Pedestrian [email protected], 0.25, 0.25:
bbox AP:59.6004, 51.0441, 42.6922
bev  AP:19.7851, 16.6067, 15.2265
3d   AP:19.1647, 15.9138, 14.6982
aos  AP:46.12, 39.23, 33.18
####Cyclist [email protected], 0.25, 0.25:
bbox AP:43.5243, 26.8833, 26.7038
bev  AP:14.7729, 10.8957, 10.6765
3d   AP:14.4502, 10.7494, 10.5691
aos  AP:32.11, 20.45, 20.40