使用对象检测 API 微调具有不同数量 类 的模型

Finetune model with different number of classes using object detection API

范围

我正在尝试使用对象检测 API 迁移学习 SSD MobileNet v3(小型)模型,但我的数据集只有 8 个 类,而提供的模型是预训练的在可可 (90 类) 上。如果我保留模型的 类 个数不变,我可以毫无问题地进行训练。

问题

更改 pipeline.config num_classes 会产生分配错误,因为图层形状与检查点变量不匹配:

tensorflow.python.framework.errors_impl.InvalidArgumentError: 2 root error(s) found.
    (0) Invalid argument: Assign requires shapes of both tensors to match. lhs shape= [1,1,288,27] rhs shape= [1,1,288,273]
        [[{{node save/Assign_15}}]]
    (1) Invalid argument: Assign requires shapes of both tensors to match. lhs shape= [1,1,288,27] rhs shape= [1,1,288,273]
        [[{{node save/Assign_15}}]]
        [[save/RestoreV2/_404]]

问题


数据集:ITD Dataset

型号:SSD MobileNetV3 - small (from the Model Zoo)

pipeline.config:

# SSDLite with Mobilenet v3 small feature extractor.
# Trained on COCO14, initialized from scratch.
# TPU-compatible.
# Users should configure the fine_tune_checkpoint field in the train config as
# well as the label_map_path and input_path fields in the train_input_reader and
# eval_input_reader. Search for "PATH_TO_BE_CONFIGURED" to find the fields that
# should be configured.

model {
  ssd {
    inplace_batchnorm_update: true
    freeze_batchnorm: false
    num_classes: 8
    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 {
      ssd_anchor_generator {
        num_layers: 6
        min_scale: 0.2
        max_scale: 0.95
        aspect_ratios: 1.0
        aspect_ratios: 2.0
        aspect_ratios: 0.5
        aspect_ratios: 3.0
        aspect_ratios: 0.3333
      }
    }
    image_resizer {
      fixed_shape_resizer {
        height: 320
        width: 320
      }
    }
    box_predictor {
      convolutional_box_predictor {
        min_depth: 0
        max_depth: 0
        num_layers_before_predictor: 0
        use_dropout: false
        dropout_keep_probability: 0.8
        kernel_size: 3
        use_depthwise: true
        box_code_size: 4
        apply_sigmoid_to_scores: false
        class_prediction_bias_init: -4.6
        conv_hyperparams {
          activation: RELU_6,
          regularizer {
            l2_regularizer {
              weight: 0.00004
            }
          }
          initializer {
            random_normal_initializer {
              stddev: 0.03
              mean: 0.0
            }
          }
          batch_norm {
            train: true,
            scale: true,
            center: true,
            decay: 0.97,
            epsilon: 0.001,
          }
        }
      }
    }
    feature_extractor {
      type: 'ssd_mobilenet_v3_small'
      min_depth: 16
      depth_multiplier: 1.0
      use_depthwise: true
      conv_hyperparams {
        activation: RELU_6,
        regularizer {
          l2_regularizer {
            weight: 0.00004
          }
        }
        initializer {
          truncated_normal_initializer {
            stddev: 0.03
            mean: 0.0
          }
        }
        batch_norm {
          train: true,
          scale: true,
          center: true,
          decay: 0.97,
          epsilon: 0.001,
        }
      }
      override_base_feature_extractor_hyperparams: true
    }
    loss {
      classification_loss {
        weighted_sigmoid_focal {
          alpha: 0.75,
          gamma: 2.0
        }
      }
      localization_loss {
        weighted_smooth_l1 {
          delta: 1.0
        }
      }
      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
        use_static_shapes: true
      }
      score_converter: SIGMOID
    }
  }
}

train_config: {
  batch_size: 16 #512
  sync_replicas: true
  startup_delay_steps: 0
  replicas_to_aggregate: 32
  num_steps: 0
  data_augmentation_options {
    ssd_random_crop_pad_fixed_aspect_ratio {
    }
  }  
  data_augmentation_options {
    random_horizontal_flip {
    }
  }
  optimizer {
    momentum_optimizer: {
      learning_rate: {
        cosine_decay_learning_rate {
          learning_rate_base: 0.4
          total_steps: 800000
          warmup_learning_rate: 0.13333
          warmup_steps: 2000
        }
      }
      momentum_optimizer_value: 0.9
    }
    use_moving_average: false
  }
  fine_tune_checkpoint: "./model/model.ckpt"
  fine_tune_checkpoint_type: "detection"
  fine_tune_checkpoint_version: V1
  keep_checkpoint_every_n_hours: 2.0
  max_number_of_boxes: 100
  unpad_groundtruth_tensors: false
}

train_input_reader: {
  tf_record_input_reader {
    input_path: "./data/train.record"
  }
  label_map_path: "./annotations/label_map.pbtxt"
  shuffle: true
}

eval_config: {
  num_examples: 1296
}

eval_input_reader: {
  tf_record_input_reader {
    input_path: "./data/val.record"
  }
  label_map_path: "./annotations/label_map.pbtxt"
  shuffle: true
  num_readers: 1
}

是的,主要是Tensorflow object detection garden对fine-tune模型的想法!你应该改变 :

fine_tune_checkpoint_type = "detection"

至:

fine_tune_checkpoint_type = "fine_tune"

然后当你调用 object_detection/model_main*.py 时,你应该注意作为参数传递的 model_dir 是空的。通过这种方式,脚本将能够加载您在配置中使用 90 类 指向的 fine_tune_checkpoint,并且它将在您的空模型目录中创建一个新的检查点,其中包含保存的权重和您的 8 类。之后,您甚至可以加载之前的自定义检查点,以防您的训练停止。

编辑:fine-tune 输入参考检查此答案:https://github.com/tensorflow/models/issues/8892#issuecomment-680207038