python - 使用 TF 1.15.5 的 model_main.py 时，评估 mAP 图很快会产生 0 并保持在那里

Question

我正在尝试为我的论文训练一个模型来检测红外 LED。我使用 TF 1.15.5 的原因是我需要将我的模型与仅支持 freeze_graph 导入的 OpenCV 一起使用。由于 TF 2 放弃了对该导出格式的支持，我必须使用 TF 1。

离开训练一夜之后，我才注意到评估的mAP只是停留在0，没有任何变化，即使现在训练已经在线8小时了。

我究竟做错了什么？该模型甚至适合我的用例吗？

显示 mAP 的图像：

显示损失的图像：

用于开始训练的命令

python model_main.py --model_dir=models/ssd_mobilenet_v2_quantized_300x300_coco_2019_01_03 --pipeline_config_path=models/ssd_mobilenet_v2_quantized_300x300_coco_2019_01_03/pipeline.config --alsologtostderr

管道配置

model {
  ssd {
    num_classes: 1
    image_resizer {
      fixed_shape_resizer {
        height: 300
        width: 300
      }
    }
    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
        }
      }
    }
    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: false
        dropout_keep_probability: 0.800000011921
        kernel_size: 1
        box_code_size: 4
        apply_sigmoid_to_scores: false
      }
    }
    anchor_generator {
      ssd_anchor_generator {
        num_layers: 6
        min_scale: 0.20000000298
        max_scale: 0.949999988079
        aspect_ratios: 1.0
        aspect_ratios: 2.0
        aspect_ratios: 0.5
        aspect_ratios: 3.0
        aspect_ratios: 0.333299994469
      }
    }
    post_processing {
      batch_non_max_suppression {
        score_threshold: 9.99999993923e-09
        iou_threshold: 0.600000023842
        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: 3000
        iou_threshold: 0.990000009537
        loss_type: CLASSIFICATION
        max_negatives_per_positive: 3
        min_negatives_per_image: 3
      }
      classification_weight: 1.0
      localization_weight: 1.0
    }
  }
}
train_config {
  batch_size: 24
  data_augmentation_options {
    random_horizontal_flip {
    }
  }
  data_augmentation_options {
    ssd_random_crop {
    }
  }
  optimizer {
    rms_prop_optimizer {
      learning_rate {
        exponential_decay_learning_rate {
          initial_learning_rate: 0.00400000018999
          decay_steps: 800720
          decay_factor: 0.949999988079
        }
      }
      momentum_optimizer_value: 0.899999976158
      decay: 0.899999976158
      epsilon: 1.0
    }
  }
  fine_tune_checkpoint: "pre-trained-models/ssd_mobilenet_v2_quantized_300x300_coco_2019_01_03/model.ckpt"
  from_detection_checkpoint: true
  num_steps: 2000000
}
train_input_reader {
  label_map_path: "annotations/tf_label_map.pbtxt"
  tf_record_input_reader {
    input_path: "annotations/train.record"
  }
}
eval_config {
  num_examples: 8000
  metrics_set: "coco_detection_metrics"
  use_moving_averages: true
  include_metrics_per_category: true
}
eval_input_reader {
  label_map_path: "annotations/tf_label_map.pbtxt"
  shuffle: false
  num_readers: 1
  tf_record_input_reader {
    input_path: "annotations/test.record"
  }
}
graph_rewriter {
  quantization {
    delay: 48000
    weight_bits: 8
    activation_bits: 8
  }
}

tf_label_map.pbtext

item {
 id: 1
 name: 'IR_LED'
}

标签： pythontensorflowopencvobject-detectionobject-detection-api