Ubuntu16.04下tensorflow+SSD实现目标检测，模型的训练（四）

tensorflow+SSD实现目标检测模型的训练

主要采用配置文件对训练模型的超参数进行配置
配置文件主要在我research/object_detection/samples/configs目录下的ssd_resent50配置文件

research/object_detection/samples/configs/ssd_resnet50_v1_fpn_shared_box_predictor_640x640_coco14_sync_face.config

下面是主要的修改后的配置文件
修改的地方有：
1.num_classes: 1
2. 填写好自己的路径
input_path: “/home/hyb/muke/models/dataset/widerface/TF-record/train.record”
label_map_path: “/home/hyb/muke/models/research/object_detection/data/face_label_map.pbtxt”
3.不需要模型预训练注释掉 #fine_tune_checkpoint: “PATH_TO_BE_CONFIGURED/model.ckpt”
4.图片大小可是256256也可以是300300图片越大获取的特征越清晰当然计算量会变大
4.batch_size：每次训练抽取多少个样本，可以是1，不宜过大，不然训练时间长

# SSD with Resnet 50 v1 FPN feature extractor, shared box predictor and focal
# loss (a.k.a Retinanet).
# See Lin et al, https://arxiv.org/abs/1708.02002
# Trained on COCO, initialized from Imagenet classification checkpoint

# Achieves 35.2 mAP on COCO14 minival dataset. Doubling the number of training
# steps to 50k gets 36.9 mAP

# This config is TPU compatible

model {
  ssd {
    inplace_batchnorm_update: true
    freeze_batchnorm: false
    num_classes: 1
    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 {
      multiscale_anchor_generator {
        min_level: 3
        max_level: 7
        anchor_scale: 4.0
        aspect_ratios: [1.0, 2.0, 0.5]
        scales_per_octave: 2
      }
    }
    image_resizer {
      fixed_shape_resizer {
        height: 256
        width: 256
      }
    }
    box_predictor {
      weight_shared_convolutional_box_predictor {
        depth: 256
        class_prediction_bias_init: -4.6
        conv_hyperparams {
          activation: RELU_6,
          regularizer {
            l2_regularizer {
              weight: 0.0004
            }
          }
          initializer {
            random_normal_initializer {
              stddev: 0.01
              mean: 0.0
            }
          }
          batch_norm {
            scale: true,
            decay: 0.997,
            epsilon: 0.001,
          }
        }
        num_layers_before_predictor: 4
        kernel_size: 3
      }
    }
    feature_extractor {
      type: 'ssd_resnet50_v1_fpn'
      fpn {
        min_level: 3
        max_level: 7
      }
      min_depth: 16
      depth_multiplier: 1.0
      conv_hyperparams {
        activation: RELU_6,
        regularizer {
          l2_regularizer {
            weight: 0.0004
          }
        }
        initializer {
          truncated_normal_initializer {
            stddev: 0.03
            mean: 0.0
          }
        }
        batch_norm {
          scale: true,
          decay: 0.997,
          epsilon: 0.001,
        }
      }
      override_base_feature_extractor_hyperparams: true
    }
    loss {
      classification_loss {
        weighted_sigmoid_focal {
          alpha: 0.25
          gamma: 2.0
        }
      }
      localization_loss {
        weighted_smooth_l1 {
        }
      }
      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: "PATH_TO_BE_CONFIGURED/model.ckpt"
  batch_size: 24
  sync_replicas: true
  startup_delay_steps: 0
  replicas_to_aggregate: 8
  num_steps: 25000
  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
    }
  }
  optimizer {
    momentum_optimizer: {
      learning_rate: {
        cosine_decay_learning_rate {
          learning_rate_base: .04
          total_steps: 25000
          warmup_learning_rate: .013333
          warmup_steps: 2000
        }
      }
      momentum_optimizer_value: 0.9
    }
    use_moving_average: false
  }
  max_number_of_boxes: 100
  unpad_groundtruth_tensors: false
}

train_input_reader: {
  tf_record_input_reader {
    input_path: "/home/hyb/muke/models/dataset/widerface/TF-record/train.record"
  }
  label_map_path: "/home/hyb/muke/models/research/object_detection/data/face_label_map.pbtxt"
}

eval_config: {
  metrics_set: "coco_detection_metrics"
  use_moving_averages: false
  num_examples: 100
}

eval_input_reader: {
  tf_record_input_reader {
    input_path: "/home/hyb/muke/models/dataset/widerface/TF-record/val.record"
  }
  label_map_path: "/home/hyb/muke/models/research/object_detection/data/face_label_map.pbtxt"
  shuffle: false
  num_readers: 1
}

配置好后在终端运行model_main.py文件

主要注意的是：
model_dir是模型训练的地址
num_train_steps是模型训练的次数我设置了9万次（但是不一定要全部训练完，我训练到模型收敛就停下来了）
alsologtostder是在控制台打印日志文件

如何判断模型收敛？
我在Tensorboard里面查看loss是否不下降了，趋于平衡就可以停止了，大概3万次左右，训练了7、8个小时这个阶段
当然得看你自己的GPU我的是英伟达2080max-q的
打开tensorboard：
在数据集的上一个目录打开，打开后在网页输入你在的地址，端口号一般是6006