# MediaPipe graph to detect/predict face landmarks on CPU.

type: "FaceLandmarkFrontCpuImage"

# Input image. (Image)
input_stream: "IMAGE:image"

# Max number of faces to detect/track. (int)
input_side_packet: "NUM_FACES:num_faces"

# Whether landmarks on the previous image should be used to help localize
# landmarks on the current image. (bool)
input_side_packet: "USE_PREV_LANDMARKS:use_prev_landmarks"

# Whether to run face mesh model with attention on lips and eyes. (bool)
# Attention provides more accuracy on lips and eye regions as well as iris
# landmarks.
input_side_packet: "WITH_ATTENTION:with_attention"

# The throttled input image. (Image)
output_stream: "IMAGE:throttled_image"
# Collection of detected/predicted faces, each represented as a list of 468 face
# landmarks. (std::vector<NormalizedLandmarkList>)
# NOTE: there will not be an output packet in the LANDMARKS stream for this
# particular timestamp if none of faces detected. However, the MediaPipe
# framework will internally inform the downstream calculators of the absence of
# this packet so that they don't wait for it unnecessarily.
output_stream: "LANDMARKS:multi_face_landmarks"

# Extra outputs (for debugging, for instance).
# Detected faces. (std::vector<Detection>)
output_stream: "DETECTIONS:face_detections"
# Regions of interest calculated based on landmarks.
# (std::vector<NormalizedRect>)
output_stream: "ROIS_FROM_LANDMARKS:face_rects_from_landmarks"
# Regions of interest calculated based on face detections.
# (std::vector<NormalizedRect>)
output_stream: "ROIS_FROM_DETECTIONS:face_rects_from_detections"

node {
  calculator: "FlowLimiterCalculator"
  input_stream: "image"
  input_stream: "FINISHED:multi_face_landmarks"
  input_stream_info: {
    tag_index: "FINISHED"
    back_edge: true
  }
  output_stream: "throttled_image"
  options: {
    [mediapipe.FlowLimiterCalculatorOptions.ext] {
      max_in_flight: 1
      max_in_queue: 1
    }
  }
}

# Converts Image to ImageFrame for FaceLandmarkFrontCpu to consume.
node {
  calculator: "FromImageCalculator"
  input_stream: "IMAGE:throttled_image"
  output_stream: "IMAGE_CPU:raw_image_frame"
  output_stream: "SOURCE_ON_GPU:is_gpu_image"
}

# TODO: Remove the extra flipping once adopting MlImage.
# If the source images are on gpu, flip the data vertically before sending them
# into FaceLandmarkFrontCpu. This maybe needed because OpenGL represents images
# assuming the image origin is at the bottom-left corner, whereas MediaPipe in
# general assumes the image origin is at the top-left corner.
node: {
  calculator: "ImageTransformationCalculator"
  input_stream: "IMAGE:raw_image_frame"
  input_stream: "FLIP_VERTICALLY:is_gpu_image"
  output_stream: "IMAGE:image_frame"
}

node {
  calculator: "FaceLandmarkFrontCpu"
  input_stream: "IMAGE:image_frame"
  input_side_packet: "NUM_FACES:num_faces"
  input_side_packet: "USE_PREV_LANDMARKS:use_prev_landmarks"
  input_side_packet: "WITH_ATTENTION:with_attention"
  output_stream: "LANDMARKS:multi_face_landmarks"
  output_stream: "DETECTIONS:face_detections"
  output_stream: "ROIS_FROM_LANDMARKS:face_rects_from_landmarks"
  output_stream: "ROIS_FROM_DETECTIONS:face_rects_from_detections"
}