[神经网络]一步一步使用Mobile-Net完成视觉识别（四）

1.环境配置

2.数据集获取

3.训练集获取

4.训练

5.调用测试训练结果

6.代码讲解

　　本文是第四篇，下载预训练模型并训练自己的数据集。

前面我们配置好了labelmap，下面我们开始下载训练好的模型。

http://download.tensorflow.org/models/object_detection/ssdlite_mobilenet_v2_coco_2018_05_09.tar.gz

下载下来解压，然后我们配置下pipeline文件

需要改动的地方有

num_classes：这个是我们的分类数，我们只有red和blue就填2

batch_size:这里我填的是2，batch_size过大，每次放入内存中训练的数据就会越多，如果你的内存不够大且数据量比较小，就填小点，我的是8G内存，图片也不过一两千张。

initial_learning_rate：学习速率，可以不修改。

fine_tune_checkpoint：输入我们下载的模型的ckpt文件的绝对路径

label_map_path：配置好的labelmap的绝对路径

tf_record_input_reader的input_path：填之前生成好的tfrecord文件的绝对路径

我的配置为以下文件：

model {
  ssd {
    num_classes: 2
    image_resizer {
      fixed_shape_resizer {
        height: 300
         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
        }
      }
      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: false
        dropout_keep_probability: 0.800000011921
        kernel_size: 3
        box_code_size: 4
        apply_sigmoid_to_scores: false
        use_depthwise: true
      }
    }
    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: 0.300000011921
        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: 2
  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: "/home/xueaoru/models/research/ssdlite_mobilenet_v2_coco_2018_05_09/model.ckpt"
  num_steps: 200000
  fine_tune_checkpoint_type: "detection"
}
train_input_reader {
  label_map_path: "/home/xueaoru/models/research/car_label_map.pbtxt"
  tf_record_input_reader {
    input_path: "/home/xueaoru/models/research/train.record"
  }
}
eval_config {
  num_examples: 60
  max_evals: 10
  use_moving_averages: false
}
eval_input_reader {
  label_map_path: "/home/xueaoru/models/research/car_label_map.pbtxt"
  shuffle: true
  num_readers: 1
  tf_record_input_reader {
    input_path: "/home/xueaoru/models/research/test.record"
  }
}

在models/research目录下执行以下命令：

python object_detection/model_main.py 
    --pipeline_config_path=/home/xueaoru/models/research/ssdlite_mobilenet_v2_coco_2018_05_09/pipeline.config 
    --num_train_steps=200000 
    --sample_1_of_n_eval_examples=25 
    --alsologtostderr 
    --model_dir=/home/xueaoru/models/research/car_data

其中pipeline_config_path为之前配置好的pipeline的绝对路径

num_train_steps为训练步数

sample_1_of_n_eval_examples为每多少个验证数据抽样一次

alsologtostderr输出std错误信息

model_dir输出训练过程中的数据的存放文件夹



执行完以上命令之后，基本上训练就开始了，我们只需要通过tensorboard来看看训练效果就可以了

tensorboard --logdir car_data

打开输出的地址：

就可以看到训练效果啦

等到差不多收敛了，我们就可以输出我们的模型了

命令行输入以下命令：

python object_detection/export_inference_graph.py --input_type image_tensor --pipeline_config_path /home/xueaoru/models/research/ssdlite_mobilenet_v2_coco_2018_05_09/pipeline.config --trained_checkpoint_prefix /home/xueaoru/models/research/car_data/model.ckpt-87564 --output_directory /home/xueaoru/models/research/inference_graph_v2

配置基本上跟上面差不多，改改路径即可。

然后我们就在inference_graph_v2目录下拿到了训练后的模型了。