chromium/third_party/mediapipe/src/mediapipe/modules/objectron/box_landmark_gpu.pbtxt

# MediaPipe Box landmark localization GPU subgraph.

type: "BoxLandmarkSubgraph"

input_stream: "IMAGE:image"
input_stream: "NORM_RECT:box_rect"
output_stream: "NORM_LANDMARKS:box_landmarks"

# Extracts image size from the input images.
node {
  calculator: "ImagePropertiesCalculator"
  input_stream: "IMAGE_GPU:image"
  output_stream: "SIZE:image_size"
}

# Expands the rectangle that contain the box so that it's likely to cover the
# entire box.
node {
  calculator: "RectTransformationCalculator"
  input_stream: "NORM_RECT:box_rect"
  input_stream: "IMAGE_SIZE:image_size"
  output_stream: "box_rect_scaled"
  options: {
    [mediapipe.RectTransformationCalculatorOptions.ext] {
      scale_x: 1.5
      scale_y: 1.5
      square_long: true
    }
  }
}

# Crops, resizes, and converts the input video into tensor.
# Preserves aspect ratio of the images.
node {
  calculator: "ImageToTensorCalculator"
  input_stream: "IMAGE_GPU:image"
  input_stream: "NORM_RECT:box_rect_scaled"
  output_stream: "TENSORS:image_tensor"
  output_stream: "LETTERBOX_PADDING:letterbox_padding"
  options {
    [mediapipe.ImageToTensorCalculatorOptions.ext] {
      output_tensor_width: 224
      output_tensor_height: 224
      keep_aspect_ratio: true
      output_tensor_float_range {
        min: 0.0
        max: 1.0
      }
      gpu_origin: TOP_LEFT
      border_mode: BORDER_REPLICATE
    }
  }
}

# Runs a TensorFlow Lite model on GPU that takes an image tensor and outputs a
# vector of tensors representing, for instance, detection boxes/keypoints and
# scores.
node {
  calculator: "InferenceCalculator"
  input_stream: "TENSORS:image_tensor"
  output_stream: "TENSORS:output_tensors"
  options: {
    [mediapipe.InferenceCalculatorOptions.ext] {
      model_path: "object_detection_3d.tflite"
      delegate { gpu {} }
    }
  }
}

# Splits a vector of tensors to multiple vectors according to the ranges
# specified in option.
node {
  calculator: "SplitTensorVectorCalculator"
  input_stream: "output_tensors"
  output_stream: "landmark_tensors"
  output_stream: "box_flag_tensor"
  options: {
    [mediapipe.SplitVectorCalculatorOptions.ext] {
      ranges: { begin: 0 end: 1 }
      ranges: { begin: 1 end: 2 }
    }
  }
}

# Converts the box-flag tensor into a float that represents the confidence
# score of box presence.
node {
  calculator: "TensorsToFloatsCalculator"
  input_stream: "TENSORS:box_flag_tensor"
  output_stream: "FLOAT:box_presence_score"
}

# Applies a threshold to the confidence score to determine whether a box is
# present.
node {
  calculator: "ThresholdingCalculator"
  input_stream: "FLOAT:box_presence_score"
  output_stream: "FLAG:box_presence"
  options: {
    [mediapipe.ThresholdingCalculatorOptions.ext] {
      threshold: 0.99
    }
  }
}

# Drops landmarks tensors if box is not present.
node {
  calculator: "GateCalculator"
  input_stream: "landmark_tensors"
  input_stream: "ALLOW:box_presence"
  output_stream: "gated_landmark_tensors"
}

# Decodes the landmark tensors into a list of landmarks, where the landmark
# coordinates are normalized by the size of the input image to the model.
node {
  calculator: "TensorsToLandmarksCalculator"
  input_stream: "TENSORS:gated_landmark_tensors"
  output_stream: "NORM_LANDMARKS:landmarks"
  options: {
    [mediapipe.TensorsToLandmarksCalculatorOptions.ext] {
      num_landmarks: 9
      input_image_width: 224
      input_image_height: 224
    }
  }
}

# Adjusts landmarks (already normalized to [0.f, 1.f]) on the letterboxed box
# image (after image transformation with the FIT scale mode) to the
# corresponding locations on the same image with the letterbox removed (box
# image before image transformation).
node {
  calculator: "LandmarkLetterboxRemovalCalculator"
  input_stream: "LANDMARKS:landmarks"
  input_stream: "LETTERBOX_PADDING:letterbox_padding"
  output_stream: "LANDMARKS:scaled_landmarks"
}

# Projects the landmarks from the cropped box image to the corresponding
# locations on the full image before cropping (input to the graph).
node {
  calculator: "LandmarkProjectionCalculator"
  input_stream: "NORM_LANDMARKS:scaled_landmarks"
  input_stream: "NORM_RECT:box_rect_scaled"
  output_stream: "NORM_LANDMARKS:box_landmarks"
}