如何在TensorFlow 2.0中正确修改目标检测模型的训练epoch数量

项目基础配置 #

我正在搭建仅用于车辆检测的模型，使用配置如下：

• 技术方案：搭载迁移学习的TensorFlow 2.0
• 预训练模型：ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8
• 数据集：仅保留car类别的coco-2017数据集

训练现象 #

训练完成后运行模型，准确率约为50%，total_loss呈下降趋势但未收敛，可查看检测车辆结果和tensorboard记录作为参考。
total_loss下降是正向表现，但模型进一步收敛可获得更高准确率。我尝试将pipeline.config中的num_epochs参数从1修改为4后没有生效，推测该参数不是调整训练轮次的正确配置项。

核心疑问 #

如何正确修改TensorFlow 2.0下模型训练的epoch数量？

现有pipeline.config配置 #

model {
  ssd {
    num_classes: 1
    image_resizer {
      fixed_shape_resizer {
        height: 320
        width: 320
      }
    }
    feature_extractor {
      type: "ssd_mobilenet_v2_fpn_keras"
      depth_multiplier: 1.0
      min_depth: 16
      conv_hyperparams {
        regularizer {
          l2_regularizer {
            weight: 4e-05
          }
        }
        initializer {
          random_normal_initializer {
            mean: 0.0
            stddev: 0.01
          }
        }
        activation: RELU_6
        batch_norm {
          decay: 0.997
          scale: true
          epsilon: 0.001
        }
      }
      use_depthwise: true
      override_base_feature_extractor_hyperparams: true
      fpn {
        min_level: 3
        max_level: 7
        additional_layer_depth: 128
      }
    }
    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 {
      }
    }
    box_predictor {
      weight_shared_convolutional_box_predictor {
        conv_hyperparams {
          regularizer {
            l2_regularizer {
              weight: 4e-05
            }
          }
          initializer {
            random_normal_initializer {
              mean: 0.0
              stddev: 0.01
            }
          }
          activation: RELU_6
          batch_norm {
            decay: 0.997
            scale: true
            epsilon: 0.001
          }
        }
        depth: 128
        num_layers_before_predictor: 4
        kernel_size: 3
        class_prediction_bias_init: -4.6
        share_prediction_tower: true
        use_depthwise: true
      }
    }
    anchor_generator {
      multiscale_anchor_generator {
        min_level: 3
        max_level: 7
        anchor_scale: 4.0
        aspect_ratios: 1.0
        aspect_ratios: 2.0
        aspect_ratios: 0.5
        scales_per_octave: 2
      }
    }
    post_processing {
      batch_non_max_suppression {
        score_threshold: 1e-08
        iou_threshold: 0.6
        max_detections_per_class: 100
        max_total_detections: 100
        use_static_shapes: false
      }
      score_converter: SIGMOID
    }
    normalize_loss_by_num_matches: true
    loss {
      localization_loss {
        weighted_smooth_l1 {
        }
      }
      classification_loss {
        weighted_sigmoid_focal {
          gamma: 2.0
          alpha: 0.25
        }
      }
      classification_weight: 1.0
      localization_weight: 1.0
    }
    encode_background_as_zeros: true
    normalize_loc_loss_by_codesize: true
    inplace_batchnorm_update: true
    freeze_batchnorm: false
  }
}
train_config {
  batch_size: 4
  data_augmentation_options {
    random_horizontal_flip {
    }
  }
  data_augmentation_options {
    random_crop_image {
      min_object_covered: 0.0
      min_aspect_ratio: 0.75
      max_aspect_ratio: 3.0
      min_area: 0.75
      max_area: 1.0
      overlap_thresh: 0.0
    }
  }
  sync_replicas: true
  optimizer {
    momentum_optimizer {
      learning_rate {
        cosine_decay_learning_rate {
          learning_rate_base: 0.0008
          total_steps: 50000
          warmup_learning_rate: 0.00026666
          warmup_steps: 1000
        }
      }
      momentum_optimizer_value: 0.9
    }
    use_moving_average: false
  }
  fine_tune_checkpoint: "Tensorflow/workspace/pre-trained-models/ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8/checkpoint/ckpt-0"
  num_steps: 50000
  startup_delay_steps: 0.0
  replicas_to_aggregate: 8
  max_number_of_boxes: 100
  unpad_groundtruth_tensors: false
  fine_tune_checkpoint_type: "detection"
  fine_tune_checkpoint_version: V2
}
train_input_reader {
  label_map_path: "Tensorflow/workspace/annotations/label_map.pbtxt"
  tf_record_input_reader {
    input_path: "Tensorflow/workspace/annotations/train.record"
  }
}
eval_config {
  metrics_set: "coco_detection_metrics"
  use_moving_averages: false
}
eval_input_reader {
  label_map_path: "Tensorflow/workspace/annotations/label_map.pbtxt"
  shuffle: false
  num_epochs: 4
  tf_record_input_reader {
    input_path: "Tensorflow/workspace/annotations/test.record"
  }
}

你之前修改的num_epochs位于eval_input_reader配置块下，该参数仅控制评估阶段测试集的遍历次数，和训练轮次无关。

TensorFlow 2.x Object Detection API默认不直接配置训练epoch数，而是通过train_config下的num_steps参数换算得到训练轮次，换算公式为：
训练轮数 = 总训练步数(num_steps) ÷ 单轮训练步数
其中单轮训练步数 = 训练集总样本数 ÷ 训练batch_size

举个实际计算示例：假设你的训练集共包含12000张标注好的车辆图片，当前配置的batch_size为4，那么单轮训练需要走12000/4=3000步；如果需要训练4轮，直接把train_config下的num_steps参数调整为3000*4=12000即可。

如果想直接指定epoch数启动训练，无需手动换算步数，也可以在启动训练的脚本（通常是model_main_tf2.py）的运行命令中添加--num_train_epochs=4参数，该参数优先级高于pipeline.config中的num_steps配置。

另外针对你当前loss未收敛的情况，除了增加训练轮次，也可以搭配调整学习率base值、补充随机缩放/色域扰动等数据增强策略，进一步提升模型准确率。

内容的提问来源于stack exchange，提问作者Hoang97