syntax = "proto2";

package object_detection.protos;

import "object_detection/protos/hyperparams.proto";

// Configuration proto for box predictor. See core/box_predictor.py for details.
message BoxPredictor {
  oneof box_predictor_oneof {
    ConvolutionalBoxPredictor convolutional_box_predictor = 1;
    MaskRCNNBoxPredictor mask_rcnn_box_predictor = 2;
    RfcnBoxPredictor rfcn_box_predictor = 3;
    WeightSharedConvolutionalBoxPredictor
        weight_shared_convolutional_box_predictor = 4;
  }
}


// Configuration proto for Convolutional box predictor.
// Next id: 13
message ConvolutionalBoxPredictor {
  // Hyperparameters for convolution ops used in the box predictor.
  optional Hyperparams conv_hyperparams = 1;

  // Minimum feature depth prior to predicting box encodings and class
  // predictions.
  optional int32 min_depth = 2 [default = 0];

  // Maximum feature depth prior to predicting box encodings and class
  // predictions. If max_depth is set to 0, no additional feature map will be
  // inserted before location and class predictions.
  optional int32 max_depth = 3 [default = 0];

  // Number of the additional conv layers before the predictor.
  optional int32 num_layers_before_predictor = 4 [default = 0];

  // Whether to use dropout for class prediction.
  optional bool use_dropout = 5 [default = true];

  // Keep probability for dropout
  optional float dropout_keep_probability = 6 [default = 0.8];

  // Size of final convolution kernel. If the spatial resolution of the feature
  // map is smaller than the kernel size, then the kernel size is set to
  // min(feature_width, feature_height).
  optional int32 kernel_size = 7 [default = 1];

  // Size of the encoding for boxes.
  optional int32 box_code_size = 8 [default = 4];

  // Whether to apply sigmoid to the output of class predictions.
  // TODO(jonathanhuang): Do we need this since we have a post processing
  // module.?
  optional bool apply_sigmoid_to_scores = 9 [default = false];

  optional float class_prediction_bias_init = 10 [default = 0.0];

  // Whether to use depthwise separable convolution for box predictor layers.
  optional bool use_depthwise = 11 [default = false];
}

// Configuration proto for weight shared convolutional box predictor.
// Next id: 18
message WeightSharedConvolutionalBoxPredictor {
  // Hyperparameters for convolution ops used in the box predictor.
  optional Hyperparams conv_hyperparams = 1;

  // Number of the additional conv layers before the predictor.
  optional int32 num_layers_before_predictor = 4 [default = 0];

  // Output depth for the convolution ops prior to predicting box encodings
  // and class predictions.
  optional int32 depth = 2 [default = 0];

  // Size of final convolution kernel. If the spatial resolution of the feature
  // map is smaller than the kernel size, then the kernel size is set to
  // min(feature_width, feature_height).
  optional int32 kernel_size = 7 [default = 3];

  // Size of the encoding for boxes.
  optional int32 box_code_size = 8 [default = 4];

  // Bias initialization for class prediction. It has been show to stabilize
  // training where there are large number of negative boxes. See
  // https://arxiv.org/abs/1708.02002 for details.
  optional float class_prediction_bias_init = 10 [default = 0.0];

  // Whether to use dropout for class prediction.
  optional bool use_dropout = 11 [default = false];

  // Keep probability for dropout.
  optional float dropout_keep_probability = 12 [default = 0.8];

  // Whether to share the multi-layer tower between box prediction and class
  // prediction heads.
  optional bool share_prediction_tower = 13 [default = false];

  // Whether to use depthwise separable convolution for box predictor layers.
  optional bool use_depthwise = 14 [default = false];

  // Enum to specify how to convert the detection scores at inference time.
  enum ScoreConverter {
    // Input scores equals output scores.
    IDENTITY = 0;

    // Applies a sigmoid on input scores.
    SIGMOID = 1;
  }

  // Callable elementwise score converter at inference time.
  optional ScoreConverter score_converter = 16 [default = IDENTITY];

  // If specified, apply clipping to box encodings.
  message BoxEncodingsClipRange {
    optional float min = 1;
    optional float max = 2;
  }
  optional BoxEncodingsClipRange box_encodings_clip_range = 17;
}

// TODO(alirezafathi): Refactor the proto file to be able to configure mask rcnn
// head easily.
// Next id: 15
message MaskRCNNBoxPredictor {
  // Hyperparameters for fully connected ops used in the box predictor.
  optional Hyperparams fc_hyperparams = 1;

  // Whether to use dropout op prior to the both box and class predictions.
  optional bool use_dropout = 2 [default = false];

  // Keep probability for dropout. This is only used if use_dropout is true.
  optional float dropout_keep_probability = 3 [default = 0.5];

  // Size of the encoding for the boxes.
  optional int32 box_code_size = 4 [default = 4];

  // Hyperparameters for convolution ops used in the box predictor.
  optional Hyperparams conv_hyperparams = 5;

  // Whether to predict instance masks inside detection boxes.
  optional bool predict_instance_masks = 6 [default = false];

  // The depth for the first conv2d_transpose op applied to the
  // image_features in the mask prediction branch. If set to 0, the value
  // will be set automatically based on the number of channels in the image
  // features and the number of classes.
  optional int32 mask_prediction_conv_depth = 7 [default = 256];

  // Whether to predict keypoints inside detection boxes.
  optional bool predict_keypoints = 8 [default = false];

  // The height and the width of the predicted mask.
  optional int32 mask_height = 9 [default = 15];
  optional int32 mask_width = 10 [default = 15];

  // The number of convolutions applied to image_features in the mask prediction
  // branch.
  optional int32 mask_prediction_num_conv_layers = 11 [default = 2];
  optional bool masks_are_class_agnostic = 12 [default = false];

  // Whether to use one box for all classes rather than a different box for each
  // class.
  optional bool share_box_across_classes = 13 [default = false];

  // Whether to apply convolutions on mask features before upsampling using
  // nearest neighbor resizing.
  // By default, mask features are resized to [`mask_height`, `mask_width`]
  // before applying convolutions and predicting masks.
  optional bool convolve_then_upsample_masks = 14 [default = false];
}

message RfcnBoxPredictor {
  // Hyperparameters for convolution ops used in the box predictor.
  optional Hyperparams conv_hyperparams = 1;

  // Bin sizes for RFCN crops.
  optional int32 num_spatial_bins_height = 2 [default = 3];

  optional int32 num_spatial_bins_width = 3 [default = 3];

  // Target depth to reduce the input image features to.
  optional int32 depth = 4 [default = 1024];

  // Size of the encoding for the boxes.
  optional int32 box_code_size = 5 [default = 4];

  // Size to resize the rfcn crops to.
  optional int32 crop_height = 6 [default = 12];

  optional int32 crop_width = 7 [default = 12];
}