Tensorflow：训练和评估过程

Question

我正在使用 Tensorflow 在本地使用 1080 Nvidia 8GB 训练我的数据集（使用对象检测 API），

我使用create_pet_tf_record.py 来生成 TFRecords 文件。我不是从头开始训练我使用mask_rcnn_inception_v2_coco_2018_01_28/model.ckpt 作为fine_tune_checkpoint。

当我运行 python object_detection/train.py 和 /eval.py 时，我会通过 Tensorboard 检查训练和评估过程。最初，一切似乎都是正确的，就像 pic1 零步长。

训练检查点间隔需要很长时间才能保存。在超过5,000 训练步骤之后，评估从/model.ckpt-0 移动到/model.ckpt-3642，整个过程此时将无法正常，如pic2 所示。

这是我的文件mask_rcnn_inception_v2.config

model {
  faster_rcnn {
    num_classes: 1
    image_resizer {
      fixed_shape_resizer {
        height: 375
        width: 500
      }
    }
    number_of_stages: 3
    feature_extractor {
      type: 'faster_rcnn_inception_v2'
      first_stage_features_stride: 16
    }
    first_stage_anchor_generator {
      grid_anchor_generator {
        scales: [0.25, 0.5, 1.0, 2.0]
        aspect_ratios: [0.5, 1.0, 2.0]
        height_stride: 16
        width_stride: 16
      }
    }
    first_stage_box_predictor_conv_hyperparams {
      op: CONV
      regularizer {
        l2_regularizer {
          weight: 0.0
        }
      }
      initializer {
        truncated_normal_initializer {
          stddev: 0.01
        }
      }
    }
    first_stage_nms_score_threshold: 0.0
    first_stage_nms_iou_threshold: 0.7
    first_stage_max_proposals: 300
    first_stage_localization_loss_weight: 2.0
    first_stage_objectness_loss_weight: 1.0
    initial_crop_size: 14
    maxpool_kernel_size: 2
    maxpool_stride: 2
    second_stage_box_predictor {
      mask_rcnn_box_predictor {
        use_dropout: false
        dropout_keep_probability: 1.0
        predict_instance_masks: true
        mask_height: 15
        mask_width: 15
        mask_prediction_conv_depth: 0
        mask_prediction_num_conv_layers: 2
        fc_hyperparams {
          op: FC
          regularizer {
            l2_regularizer {
              weight: 0.0
            }
          }
          initializer {
            variance_scaling_initializer {
              factor: 1.0
              uniform: true
              mode: FAN_AVG
            }
          }
        }
        conv_hyperparams {
          op: CONV
          regularizer {
            l2_regularizer {
              weight: 0.0
            }
          }
          initializer {
            truncated_normal_initializer {
              stddev: 0.01
            }
          }
        }
      }
    }
    second_stage_post_processing {
      batch_non_max_suppression {
        score_threshold: 0.0
        iou_threshold: 0.6
        max_detections_per_class: 100
        max_total_detections: 300
      }
      score_converter: SOFTMAX
    }
    second_stage_localization_loss_weight: 2.0
    second_stage_classification_loss_weight: 1.0
    second_stage_mask_prediction_loss_weight: 4.0
  }
}

train_config: {
  batch_size: 1
  optimizer {
    momentum_optimizer: {
      learning_rate: {
        manual_step_learning_rate {
          initial_learning_rate: 0.0002
          schedule {
            step: 900000
            learning_rate: .00002
          }
          schedule {
            step: 1200000
            learning_rate: .000002
          }
        }
      }
      momentum_optimizer_value: 0.9
    }
    use_moving_average: false
  }
gradient_clipping_by_norm: 10.0

  fine_tune_checkpoint: "/home/jesse/gpu-py3/models/research/object_detection/models/model/mask_rcnn_inception_v2_coco_train/mask_rcnn_inception_v2_coco_2018_01_28/model.ckpt"
  from_detection_checkpoint: true
  # Note: The below line limits the training process to 200K steps, which we
  # empirically found to be sufficient enough to train the pets dataset. This
  # effectively bypasses the learning rate schedule (the learning rate will
  # never decay). Remove the below line to train indefinitely.
  num_steps: 200000
  data_augmentation_options {
    random_horizontal_flip {
    }
  }
}

train_input_reader: {
  tf_record_input_reader {
    input_path: "/home/jesse/gpu-py3/models/research/ttt/pet_train.record"
  }
  label_map_path: "/home/jesse/gpu-py3/models/research/object_detection/data/pet_label_map.pbtxt"
  load_instance_masks: true
  mask_type: PNG_MASKS
}

eval_config: {
  num_examples: 8000
  # Note: The below line limits the evaluation process to 10 evaluations.
  # Remove the below line to evaluate indefinitely.
  max_evals: 10
}

eval_input_reader: {
  tf_record_input_reader {
    input_path: "/home/jesse/gpu-py3/models/research/ttt/pet_val.record"
  }
  label_map_path: "/home/jesse/gpu-py3/models/research/object_detection/data/pet_label_map.pbtxt"
  load_instance_masks: true
  mask_type: PNG_MASKS
  shuffle: false
  num_readers: 1
}

我不知道我错在哪里，我觉得我应该更频繁地运行评估，例如应该每 2000 步保存一次训练检查点。或者我可能需要编辑管道文件mask_rcnn_inception_v2.config。我不知道为什么在pic2 中看到的3642 步骤之后训练结果非常失望。

非常感谢任何帮助

Answer 1

我的 2 美分，假设您没有对重要的配置参数进行太多修改，您的训练数据非常多样化，并且随着更多的迭代通过其泛化。尝试更准确地标记图像，即使这意味着更少的图像。