def evaluate()

in tensorflow_graphics/projects/points_to_3Dobjects/utils/evaluator.py [0:0]

• 52 lines of code
• 15 McCabe index (conditional complexity)

  def evaluate(self):
    """Eval."""
    predictions_per_class = {}  # map {classname: pred}
    labels_per_class = {}  # map {classname: gt}

    for scene_id in self.predicted_boxes:
      bboxes, classnames, scores = self.predicted_boxes[scene_id]
      classnames = classnames.numpy()
      bboxes = bboxes.numpy()
      scores = scores.numpy()
      for i in range(classnames.shape[0]):
        classname = classnames[i]
        bbox = bboxes[i]
        score = scores[i]
        # for classname, bbox, score in self.predicted_boxes[scene_id]:
        if classname not in predictions_per_class:
          predictions_per_class[classname] = {}
        if scene_id not in predictions_per_class[classname]:
          predictions_per_class[classname][scene_id] = []
        if classname not in labels_per_class:
          labels_per_class[classname] = {}
        if scene_id not in labels_per_class[classname]:
          labels_per_class[classname][scene_id] = []
        predictions_per_class[classname][scene_id].append((bbox, score))

    for scene_id in self.labeled_boxes:
      bboxes, classnames = self.labeled_boxes[scene_id]
      classnames = classnames.numpy()
      bboxes = bboxes.numpy()
      for i in range(classnames.shape[0]):
        classname = classnames[i]
        bbox = bboxes[i]
        if classname not in labels_per_class:
          labels_per_class[classname] = {}
        if scene_id not in labels_per_class[classname]:
          labels_per_class[classname][scene_id] = []
        labels_per_class[classname][scene_id].append(bbox)

    recall_per_class = {}
    precision_per_class = {}
    ap_per_class = {}
    for classname in labels_per_class:
      print('Computing AP for class: ', classname)
      if classname in predictions_per_class:
        recall, precision, ap = self._eval_detections_per_class(
            # this does not work when class was never predicted
            predictions_per_class[classname],
            labels_per_class[classname],
            self.threshold)
      else:
        recall, precision, ap = 0.0, 0.0, 0.0
      recall_per_class[classname] = recall
      precision_per_class[classname] = precision
      ap_per_class[classname] = ap
      print(classname, ap)
    # return recall_per_class, precision_per_class, ap_per_class
    mean = np.mean(np.array([v for k, v in ap_per_class.items()]))
    print(mean)
    return mean