tensorflow/models

# SpaghettiNet Feature Extractor optimized for EdgeTPU.
# Trained on COCO17 from scratch.
#
# spaghettinet_edgetpu_s
# Achieves 26.2% mAP on COCO17 at 400k steps.
# 1.31ms Edge TPU latency at 1 billion MACs, 3.4 million params.
#
# spaghettinet_edgetpu_m
# Achieves 27.4% mAP on COCO17 at 400k steps.
# 1.55ms Edge TPU latency at 1.25 billion MACs, 4.1 million params.
#
# spaghettinet_edgetpu_l
# Achieves 28.02% mAP on COCO17 at 400k steps.
# 1.75ms Edge TPU latency at 1.57 billion MACs, 5.7 million params.
#
# TPU-compatible.

model {
  ssd {
    inplace_batchnorm_update: true
    freeze_batchnorm: false
    num_classes: 90
    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: 5
        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: 0.3333333
      }
    }
    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_spaghettinet'
      # 3 architectures are supported and performance for each is listed at the top of this config file.
      #spaghettinet_arch_name: 'spaghettinet_edgetpu_s'
      spaghettinet_arch_name: 'spaghettinet_edgetpu_m'
      #spaghettinet_arch_name: 'spaghettinet_edgetpu_l'
      use_explicit_padding: false
    }
    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: 512
  sync_replicas: true
  startup_delay_steps: 0
  replicas_to_aggregate: 32
  num_steps: 400000
  data_augmentation_options {
    random_horizontal_flip {
    }
  }
  data_augmentation_options {
    ssd_random_crop {
    }
  }
  optimizer {
    momentum_optimizer: {
      learning_rate: {
        cosine_decay_learning_rate {
          learning_rate_base: 0.8
          total_steps: 400000
          warmup_learning_rate: 0.13333
          warmup_steps: 2000
        }
      }
      momentum_optimizer_value: 0.9
    }
    use_moving_average: false
  }
  max_number_of_boxes: 100
  unpad_groundtruth_tensors: false
}

train_input_reader: {
  label_map_path: "PATH_TO_BE_CONFIGURED/label_map.txt"
  tf_record_input_reader {
    input_path: "PATH_TO_BE_CONFIGURED/train2017-?????-of-00256.tfrecord"
  }
}

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

eval_input_reader: {
  label_map_path: "PATH_TO_BE_CONFIGURED/label_map.txt"
  shuffle: false
  num_epochs: 1
  tf_record_input_reader {
    input_path: "PATH_TO_BE_CONFIGURED/val2017-?????-of-00032.tfrecord"
  }
}

graph_rewriter {
  quantization {
    delay: 40000
    weight_bits: 8
    activation_bits: 8
  }
}