Object Detection Confidence Distribution

Question

Im training a MobilenetSSD_v2 object detector (pretrained on COCO dataset) using TensorFlow Object Detection API. The model is working ok, however the confidences of the output boxes is pretty weird:

Here is my config file:

model {  ssd {
num_classes: 5
image_resizer {
  fixed_shape_resizer {
    height: 270
    width: 480
  }
}
  feature_extractor {
  type: "ssd_mobilenet_v2"
  depth_multiplier: 1.0
  min_depth: 16
  conv_hyperparams {
    regularizer {
      l2_regularizer {
        weight: 3.99999989895e-05
      }
    }
    initializer {
      truncated_normal_initializer {
        mean: 0.0
        stddev: 0.0299999993294
      }
    }
    activation: RELU_6
    batch_norm {
      decay: 0.999700009823
      center: true
      scale: true
      epsilon: 0.0010000000475
      train: true
    }
  }
  use_depthwise: true
}
box_coder {
  faster_rcnn_box_coder {
    y_scale: 10.0
    x_scale: 10.0
    height_scale: 5.0
    width_scale: 5.0
  }
}
matcher {
  argmax_matcher {
    matched_threshold: 0.5
    unmatched_threshold: 0.5
    ignore_thresholds: false
    negatives_lower_than_unmatched: true
    force_match_for_each_row: true
  }
}
similarity_calculator {
  iou_similarity {
  }
}
box_predictor {
  convolutional_box_predictor {
    conv_hyperparams {
      regularizer {
        l2_regularizer {
          weight: 3.99999989895e-05
        }
      }
      initializer {
        truncated_normal_initializer {
          mean: 0.0
          stddev: 0.0299999993294
        }
      }
      activation: RELU_6
      batch_norm {
        decay: 0.999700009823
        center: true
        scale: true
        epsilon: 0.0010000000475
        train: true
      }
    }
    min_depth: 0
    max_depth: 0
    num_layers_before_predictor: 0
    use_dropout: true
    dropout_keep_probability: 0.800000011921
    kernel_size: 3
    box_code_size: 4
    apply_sigmoid_to_scores: false
  }
}
anchor_generator {
  ssd_anchor_generator {
    num_layers: 6
    min_scale: 0.15
    max_scale: 0.949999988079
    aspect_ratios: 0.1
    aspect_ratios: 0.2
    aspect_ratios: 0.5
    aspect_ratios: 1.0
    aspect_ratios: 2.0
    aspect_ratios: 5.0
    aspect_ratios: 10.0
    
  }
}
post_processing {
  batch_non_max_suppression {
    score_threshold: 0.4
    iou_threshold: 0.000001
    max_detections_per_class: 100
    max_total_detections: 100
  }
  score_converter: SIGMOID
}
normalize_loss_by_num_matches: true
loss {
  localization_loss {
    weighted_smooth_l1 {
    }
  }
  classification_loss {
    weighted_sigmoid {
    }
  }
  hard_example_miner {
    num_hard_examples: 10
    iou_threshold: 0.7
    loss_type: CLASSIFICATION
    max_negatives_per_positive: 1
    min_negatives_per_image: 1
  }
  classification_weight: 0.9
  localization_weight: 1.0
}  }}
train_config {
  batch_size: 32
  optimizer {
    adam_optimizer {
      learning_rate {
        exponential_decay_learning_rate {
          initial_learning_rate: 0.0003
          decay_steps: 1000
          decay_factor: 0.95
        }
      }
    }
     use_moving_average: false
  }

 fine_tune_checkpoint: "ssd_mobilenet_v2_coco/model.ckpt"
 num_steps: 10000
 fine_tune_checkpoint_type: "detection"
}...

Does anyone have an idea what can couse such behaviour? And why doesnt i get values smaller then 0.4, or higher then 0.6?