chromium/third_party/mediapipe/src/mediapipe/graphs/object_detection/object_detection_desktop_tensorflow_graph.pbtxt

# MediaPipe graph that performs object detection on desktop with TensorFlow
# on CPU.
# Used in the example in
# mediapipie/examples/desktop/object_detection:object_detection_tensorflow.

# Decodes an input video file into images and a video header.
node {
  calculator: "OpenCvVideoDecoderCalculator"
  input_side_packet: "INPUT_FILE_PATH:input_video_path"
  output_stream: "VIDEO:input_video"
  output_stream: "VIDEO_PRESTREAM:input_video_header"
}

# Converts the input image into an image tensor as a tensorflow::Tensor.
node {
  calculator: "ImageFrameToTensorCalculator"
  input_stream: "input_video"
  output_stream: "image_tensor"
}

# Generates a single side packet containing a TensorFlow session from a saved
# model. The directory path that contains the saved model is specified in the
# saved_model_path option, and the name of the saved model file has to be
# "saved_model.pb".
node {
  calculator: "TensorFlowSessionFromSavedModelCalculator"
  output_side_packet: "SESSION:object_detection_session"
  node_options: {
    [type.googleapis.com/mediapipe.TensorFlowSessionFromSavedModelCalculatorOptions]: {
      saved_model_path: "mediapipe/models/object_detection_saved_model"
    }
  }
}

# Runs a TensorFlow session (specified as an input side packet) that takes an
# image tensor and outputs multiple tensors that describe the objects detected
# in the image. The batch_size option is set to 1 to disable batching entirely.
# Note that the particular TensorFlow model used in this session handles image
# scaling internally before the object-detection inference, and therefore no
# additional calculator for image transformation is needed in this MediaPipe
# graph.
node: {
  calculator: "TensorFlowInferenceCalculator"
  input_side_packet: "SESSION:object_detection_session"
  input_stream: "INPUTS:image_tensor"
  output_stream: "DETECTION_BOXES:detection_boxes_tensor"
  output_stream: "DETECTION_CLASSES:detection_classes_tensor"
  output_stream: "DETECTION_SCORES:detection_scores_tensor"
  output_stream: "NUM_DETECTIONS:num_detections_tensor"
  node_options: {
    [type.googleapis.com/mediapipe.TensorFlowInferenceCalculatorOptions]: {
      batch_size: 1
    }
  }
}

# Decodes the detection tensors from the TensorFlow model into a vector of
# detections. Each detection describes a detected object.
node {
  calculator: "ObjectDetectionTensorsToDetectionsCalculator"
  input_stream: "BOXES:detection_boxes_tensor"
  input_stream: "SCORES:detection_scores_tensor"
  input_stream: "CLASSES:detection_classes_tensor"
  input_stream: "NUM_DETECTIONS:num_detections_tensor"
  output_stream: "DETECTIONS:detections"
}

# Performs non-max suppression to remove excessive detections.
node {
  calculator: "NonMaxSuppressionCalculator"
  input_stream: "detections"
  output_stream: "filtered_detections"
  node_options: {
    [type.googleapis.com/mediapipe.NonMaxSuppressionCalculatorOptions] {
      min_suppression_threshold: 0.4
      min_score_threshold: 0.6
      max_num_detections: 10
      overlap_type: INTERSECTION_OVER_UNION
    }
  }
}

# Maps detection label IDs to the corresponding label text. The label map is
# provided in the label_map_path option.
node {
  calculator: "DetectionLabelIdToTextCalculator"
  input_stream: "filtered_detections"
  output_stream: "output_detections"
  node_options: {
    [type.googleapis.com/mediapipe.DetectionLabelIdToTextCalculatorOptions] {
      label_map_path: "mediapipe/models/ssdlite_object_detection_labelmap.txt"
    }
  }
}

# Converts the detections to drawing primitives for annotation overlay.
node {
  calculator: "DetectionsToRenderDataCalculator"
  input_stream: "DETECTIONS:output_detections"
  output_stream: "RENDER_DATA:render_data"
  node_options: {
    [type.googleapis.com/mediapipe.DetectionsToRenderDataCalculatorOptions] {
      thickness: 4.0
      color { r: 255 g: 0 b: 0 }
    }
  }
}

# Draws annotations and overlays them on top of the input images.
node {
  calculator: "AnnotationOverlayCalculator"
  input_stream: "IMAGE:input_video"
  input_stream: "render_data"
  output_stream: "IMAGE:output_video"
}

# Encodes the annotated images into a video file, adopting properties specified
# in the input video header, e.g., video framerate.
node {
  calculator: "OpenCvVideoEncoderCalculator"
  input_stream: "VIDEO:output_video"
  input_stream: "VIDEO_PRESTREAM:input_video_header"
  input_side_packet: "OUTPUT_FILE_PATH:output_video_path"
  node_options: {
    [type.googleapis.com/mediapipe.OpenCvVideoEncoderCalculatorOptions]: {
      codec: "avc1"
      video_format: "mp4"
    }
  }
}