如何获得带有自定义数据集的Faster R-CNN模型以实际检测对象?

时间:2019-05-05 11:27:05

标签: python tensorflow machine-learning object-detection object-detection-api

我将带有Pascal VOC样式标注的数据集转换为TFRecord文件格式,并尝试使用Tensorflow配置修改后的版本在fast_rcnn_inception_v2_coco_2018_01_28.tar.gz

中训练Faster R-CNN

但是,当运行model_main.py时,我得到警告,几乎所有的班级“都没有地面真理的例子” (我已经在这里发布了此内容:How to fix "the following classes have no ground truth examples" when running object_detection/model_main.py?) 张量板在右侧图像中显示了正确的地面真相,但在左侧图像中没有检测到。

为什么我的模型没有做出任何预测并且mAP = 0?也许问题是我没有使用https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md中的模型 已经以Pascal mAP为指标?


这是我的模型.config文件(当然,路径正确):

model {
  faster_rcnn {
    num_classes: 821
    image_resizer {
      keep_aspect_ratio_resizer {
        min_dimension: 600
        max_dimension: 1024
      }
    }
    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
        fc_hyperparams {
          op: FC
          regularizer {
            l2_regularizer {
              weight: 0.0
            }
          }
          initializer {
            variance_scaling_initializer {
              factor: 1.0
              uniform: true
              mode: FAN_AVG
            }
          }
        }
      }
    }
    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
  }
}
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: "PATH_TO/models/model/model.ckpt"
  #from_detection_checkpoint: true
  #load_all_detection_checkpoint_vars: true
  num_steps: 5000

  data_augmentation_options {
    random_horizontal_flip {
    }
  }
  data_augmentation_options {
    random_vertical_flip {
    }
  }
  data_augmentation_options {
    random_rotation90 {
    }
  }
}

train_input_reader {
  label_map_path: "PATH_TO/data/pascal_label_map.pbtxt"
  tf_record_input_reader {
    input_path:"PATH_TO/data/pascal_train.record-?????-of-00010"
  }
}
eval_config {
  num_examples: 1886
  # Note: The below line limits the evaluation process to 100     evaluations.
  # Remove the below line to evaluate indefinitely.
  max_evals: 1886
  #use_moving_averages: false
  metrics_set: "pascal_voc_detection_metrics"
}
eval_input_reader {
  label_map_path:  "PATH_TO/data/pascal_label_map.pbtxt"
  shuffle: false
  num_readers: 10
  tf_record_input_reader {
    input_path: "PATH_TO/data/pascal_val.record-?????-of-00010"
  }
}

0 个答案:

没有答案