张量板损失函数中y轴的单位是什么？

Question

我正在通过 Tensorflow 训练模型并通过 Tensorboard 进行评估。这是我的总损失函数：

谁能告诉我y轴的单位是什么？一开始我以为它是一个比例，但你不会期望它从 > 4 开始。我知道这是分类损失和定位损失的组合，但即使是单独的分类损失也从 > 3 开始。

我正在通过终端命令进行训练：

set NVIDIA_VISIBLE_DEVICES=0 & set CUDA_VISIBLE_DEVICES=0 & python object_detection/model_main_tf2.py --pipeline_config_path="V:/Projecten/A70_30_65/Marterkist/Model/ssd_mobilenet_v2_320x320_coco17_tpu-8.config" --model_dir="V:/Projecten/A70_30_65/Marterkist/Training" --alsologtostderr

并通过终端命令进行评估：

python object_detection/model_main_tf2.py --pipeline_config_path="V:/Projecten/A70_30_65/Marterkist/Model/ssd_mobilenet_v2_320x320_coco17_tpu-8.config" --model_dir="V:/Projecten/A70_30_65/Marterkist/Training" --checkpoint_dir="V:/Projecten/A70_30_65/Marterkist/Training" --alsologtostderr

这是关联的.config 文件：

# SSD with Mobilenet v2
# Trained on COCO17, initialized from Imagenet classification checkpoint
# Train on TPU-8
#
# Achieves 22.2 mAP on COCO17 Val

model {
  ssd {
    inplace_batchnorm_update: true
    freeze_batchnorm: false
    num_classes: 7
    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
        use_matmul_gather: true
      }
    }
    similarity_calculator {
      iou_similarity {
      }
    }
    encode_background_as_zeros: true
    anchor_generator {
      ssd_anchor_generator {
        num_layers: 6
        min_scale: 0.2
        max_scale: 0.95
        aspect_ratios: 1.0
        aspect_ratios: 2.0
        aspect_ratios: 0.5
        aspect_ratios: 3.0
        aspect_ratios: 0.3333
      }
    }
    image_resizer {
      fixed_shape_resizer {
        height: 300
        width: 300
      }
    }
    box_predictor {
      convolutional_box_predictor {
        min_depth: 0
        max_depth: 0
        num_layers_before_predictor: 0
        use_dropout: false
        dropout_keep_probability: 0.8
        kernel_size: 1
        box_code_size: 4
        apply_sigmoid_to_scores: false
        class_prediction_bias_init: -4.6
        conv_hyperparams {
          activation: RELU_6,
          regularizer {
            l2_regularizer {
              weight: 0.00004
            }
          }
          initializer {
            random_normal_initializer {
              stddev: 0.01
              mean: 0.0
            }
          }
          batch_norm {
            train: true,
            scale: true,
            center: true,
            decay: 0.97,
            epsilon: 0.001,
          }
        }
      }
    }
    feature_extractor {
      type: 'ssd_mobilenet_v2_keras'
      min_depth: 16
      depth_multiplier: 1.0
      conv_hyperparams {
        activation: RELU_6,
        regularizer {
          l2_regularizer {
            weight: 0.00004
          }
        }
        initializer {
          truncated_normal_initializer {
            stddev: 0.03
            mean: 0.0
          }
        }
        batch_norm {
          train: true,
          scale: true,
          center: true,
          decay: 0.97,
          epsilon: 0.001,
        }
      }
      override_base_feature_extractor_hyperparams: true
    }
    loss {
      classification_loss {
        weighted_sigmoid_focal {
          alpha: 0.75,
          gamma: 2.0
        }
      }
      localization_loss {
        weighted_smooth_l1 {
          delta: 1.0
        }
      }
      classification_weight: 1.0
      localization_weight: 1.0
    }
    normalize_loss_by_num_matches: true
    normalize_loc_loss_by_codesize: true
    post_processing {
      batch_non_max_suppression {
        score_threshold: 1e-8
        iou_threshold: 0.6
        max_detections_per_class: 100
        max_total_detections: 100
      }
      score_converter: SIGMOID
    }
  }
}

train_config: {
  fine_tune_checkpoint_version: V2
  fine_tune_checkpoint: "V:/Projecten/A70_30_65/Marterkist/Model/ssd_mobilenet_v2_320x320_coco17_tpu-8/checkpoint/ckpt-0"
  fine_tune_checkpoint_type: "detection"
  batch_size: 32
  sync_replicas: true
  startup_delay_steps: 0
  replicas_to_aggregate: 8
  num_steps: 25000

  optimizer {
    momentum_optimizer: {
      learning_rate: {
        manual_step_learning_rate {
          initial_learning_rate: 0.0003
          schedule {
            step: 20000
            learning_rate: 0.0003
          }
        }
      }
      momentum_optimizer_value: 0.9
    }
    use_moving_average: false
  }
  max_number_of_boxes: 100
  unpad_groundtruth_tensors: false
}

train_input_reader: {
  label_map_path: "V:/Projecten/A70_30_65/Marterkist/Model/labelmap.pbtxt"
  tf_record_input_reader {
    input_path: "V:/Projecten/A70_30_65/Marterkist/Data/Train/train.record"
  }
}

eval_config: {
  metrics_set: "coco_detection_metrics"
  use_moving_averages: false
}

eval_input_reader: {
  label_map_path: "V:/Projecten/A70_30_65/Marterkist/Model/labelmap.pbtxt"
  shuffle: false
  num_epochs: 1
  tf_record_input_reader {
    input_path: "V:/Projecten/A70_30_65/Marterkist/Data/Train/test.record"
  }
}

Answer 1

你的配置的相关部分是这样的：

    loss {
      classification_loss {
        weighted_sigmoid_focal {
          alpha: 0.75,
          gamma: 2.0
        }
      }
      localization_loss {
        weighted_smooth_l1 {
          delta: 1.0
        }
      }
      classification_weight: 1.0
      localization_weight: 1.0
    }

classification_weight = localization_weight = 1 表示总损失只是分类损失和定位损失的总和。 weighted_sigmoid_focal分类损失计算为-alpha*(1 - p)**gamma*log(p)，其中p是类概率（详见https://www.tensorflow.org/addons/api_docs/python/tfa/losses/SigmoidFocalCrossEntropy引用的文章）。很难赋予它一些易于解释的意义。而weighted_smooth_l1定位损失与Huber loss相同，也不易解释。

以上所有内容归结为以下几点：您看到的绝对值没有任何易于理解的含义。重要的是相对变化：损失是增加还是减少等等。