Overview of Magritte Graphs

Top-level graphs
Subgraphs

This page gives an overview of all Magritte top-level and feature graphs. See the concepts page for an explanation of this terminology.

Top-level graphs

FaceBlurWithTrackingLiveCpu

A graph that detects/tracks faces and blurs them using simple box blur.

Note that simple blurring is not an effective de-identification method!

The face detection supports all orientations and both short and full ranges.

Once detected, moving faces are tracked with mediapipe object tracking.

This graph is specialized for CPU architectures and live environments, tracking movements at a fixed frame rate of 5 fps.

Input streams:

input_video: An ImageFrame stream containing the image on which detection models are run.

Output streams:

output_video: An ImageFrame stream containing the blurred image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_blur_with_tracking_live_cpu

Text proto file:

@magritte//magritte/graphs:face_blur_with_tracking_live_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_blur_with_tracking_live_cpu_graph

Code: source code

FaceBlurWithTrackingOfflineCpu

A graph that detects/tracks faces and blurs them using simple box blur.

Note that simple blurring is not an effective de-identification method!

The face detection supports all orientations and both short and full ranges.

Once detected, moving faces are tracked with mediapipe object tracking.

This graph is specialized for CPU architectures and offline environments, tracking movements across all frames.

Input streams:

input_video: An ImageFrame stream containing the image on which detection models are run.

Output streams:

output_video: An ImageFrame stream containing the blurred image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_blur_with_tracking_offline_cpu

Text proto file:

@magritte//magritte/graphs:face_blur_with_tracking_offline_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_blur_with_tracking_offline_cpu_graph

Code: source code

FaceOverlayLiveGpu

A graph that detects faces and draws debug information.

It draws in red the raw detection output (detection box, keypoints, score), and in green the redacted area outline.

This graph is specialized for GPU architectures. It also optimized for live streaming environments by throttling the input stream and by applying only full-range face detection.

Input streams:

input_video: A GpuBuffer stream containing the image on which detection models are run.

Output streams:

output_video: A GpuBuffer stream containing the image annotated with detection and redaction areas.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_overlay_live_gpu

Text proto file:

@magritte//magritte/graphs:face_overlay_live_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_overlay_live_gpu_graph

Code: source code

FaceOverlayOfflineCpu

A graph that detects faces and draws debug information.

It draws in red the raw detection output (detection box, keypoints, score), and in green the redacted area outline.

This graph is specialized for CPU architectures and offline environments (no throttling is applied).

Input streams:

input_video: An ImageFrame stream containing the image on which detection models are run.

Output streams:

output_video: An ImageFrame stream containing the image annotated with detection and redaction areas.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_overlay_offline_cpu

Text proto file:

@magritte//magritte/graphs:face_overlay_offline_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_overlay_offline_cpu_graph

Code: source code

FacePixelizationLiveGpu

A graph that detects and redacts faces by pixelizing them.

This graph is specialized for GPU architectures. It also optimized for live streaming environments by throttling the input stream and by applying only full-range face detection.

Input streams:

input_video: The GpuBuffer stream containing the image to be redacted.

Output streams:

output_video: A GpuBuffer stream containing the redacted image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_pixelization_live_gpu

Text proto file:

@magritte//magritte/graphs:face_pixelization_live_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_pixelization_live_gpu_graph

Code: source code

FacePixelizationOfflineCpu

A graph that detects and redacts faces by pixelizing them.

This graph is specialized for CPU architectures and offline environments (no throttling is applied).

Input streams:

input_video: The ImageFrame stream containing the image to be redacted.

Output streams:

output_video: An ImageFrame stream containing the redacted image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_pixelization_offline_cpu

Text proto file:

@magritte//magritte/graphs:face_pixelization_offline_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_pixelization_offline_cpu_graph

Code: source code

FaceStickerRedactionLiveGpu

A graph that detects and redacts faces with an opaque “sticker” image.

The sticker image file is specified manually in this graph. We use one of the files in the test_data directory.

This graph is specialized for GPU architectures. It also optimized for live streaming environments by throttling the input stream and by applying only full-range face detection.

Input streams:

input_video: The GpuBuffer stream containing the image to be redacted.

Output streams:

output_video: A GpuBuffer stream containing the redacted image.

Input side packets:

sticker_image: The file path for the sticker image in the PNG format. If not provided, defaults to emoji.png.
sticker_zoom: The default sticker zoom as a float. This is a default extra zoom that is applied to the sticker once it covers the face detection bounding box. If not provided, the default value is 1.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_sticker_redaction_live_gpu

Text proto file:

@magritte//magritte/graphs:face_sticker_redaction_live_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_sticker_redaction_live_gpu_graph

Code: source code

FaceStickerRedactionOfflineCpu

A graph that detects and redacts faces with an opaque “sticker” image.

The sticker image file is specified manually in this graph. We use one of the files in the test_data directory.

Input streams:

input_video: The ImageFrame stream containing the image to be redacted.

Output streams:

output_video: An ImageFrame stream containing the redacted image.

Input side packets:

sticker_image: The file path for the sticker image in the PNG format. If not provided, defaults to emoji.png.
sticker_zoom: The default sticker zoom as a float. This is a default extra zoom that is applied to the sticker once it covers the face detection bounding box. If not provided, the default value is 1.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_sticker_redaction_offline_cpu

Text proto file:

@magritte//magritte/graphs:face_sticker_redaction_offline_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_sticker_redaction_offline_cpu_graph

Code: source code

FaceTrackingOverlayLiveGpu

A graph that detects and tracks faces to draw debug information.

It draws in red the raw detection output (detection box, keypoints, score), and in blue the tracked detection output (detection box, keypoints, score, when present).

This graph is specialized for GPU architectures. It also optimized for live streaming environments by throttling the input stream and by applying only full-range face detection.

Input streams:

input_video: A GpuBuffer stream containing the image on which detection models are run.

Output streams:

output_video: A GpuBuffer stream containing the image annotated with detections data.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_tracking_overlay_live_gpu

Text proto file:

@magritte//magritte/graphs:face_tracking_overlay_live_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_tracking_overlay_live_gpu_graph

Code: source code

FaceTrackingOverlayOfflineCpu

A graph that detects and tracks faces to draw debug information.

It draws in red the raw detection output (detection box, keypoints, score), and in blue the tracked detection output (detection box, keypoints, score, when present).

This graph is specialized for CPU architectures and offline environments (no throttling is applied).

Input streams:

input_video: An ImageFrame stream containing the image on which detection models are run.

Output streams:

output_video: An ImageFrame stream containing the image annotated with detections data.

Build targets:

Graph cc_library:

@magritte//magritte/graphs:face_tracking_overlay_offline_cpu

Text proto file:

@magritte//magritte/graphs:face_tracking_overlay_offline_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs:face_tracking_overlay_offline_cpu_graph

Code: source code

Subgraphs

Detection

FaceDetection360ShortAndFullRangeSubgraphCpu

A face detection subgraph that supports all orientations, and both short and full ranges.

This subgraph utilizes a face detection method that only supports orientations of up to +/- 45°. Thus the image is rotated 0°, 90°, 180° and 270°, the detection method is applied in each rotated image, and the corresponding detections are rotated back. Finally, a non-max suppression is applied to remove duplicate detections.

Input streams:

IMAGE: The ImageFrame stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_360_short_and_full_range_cpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_360_short_and_full_range_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_360_short_and_full_range_cpu_graph

Code: source code

FaceDetection360ShortAndFullRangeSubgraphGpu

A face detection subraph that supports all orientations, and both short and full ranges.

Input streams:

IMAGE: The GpuBuffer stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_360_short_and_full_range_gpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_360_short_and_full_range_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_360_short_and_full_range_gpu_graph

Code: source code

FaceDetection360ShortRangeByRoiSubgraphCpu

A face detection subgraph that supports all orientations.

This subgraph only supports short-range detection, e.g. from a phone’s front camera.

This subgraph utilizes a face detection method that only supports orientations of up to +/- 45°, but can take a Region of Interest (ROI) as a NormalizedRect, on which the faces will be detected. Thus, a ROI including the whole image is created for rotations of 0°, 90°, 180° and 270°, and the detection method is applied in those ROIs. Finally, a non-max suppression is applied to remove duplicate detections.

This subgraph is up to 5% more efficient than FaceDetection360ShortAndFullRangeSubgraphCpu, though it does not support full-range detection.

Input streams:

IMAGE: The ImageFrame stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_360_short_range_by_roi_cpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_360_short_range_by_roi_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_360_short_range_by_roi_cpu_graph

Code: source code

FaceDetection360ShortRangeByRoiSubgraphGpu

A face detection subgraph that supports all orientations.

This subgraph only supports short-range detection, e.g. from a phone’s front camera.

This subgraph is up to 5% more efficient than FaceDetection360ShortAndFullRangeSubgraphGpu, though it does not support full-range detection.

Input streams:

IMAGE: The GpuBuffer stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_360_short_range_by_roi_gpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_360_short_range_by_roi_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_360_short_range_by_roi_gpu_graph

Code: source code

FaceDetectionFullRangeSubgraphCpu

A full-range face detection subgraph.

This subgraph only supports full-range detection, e.g. from a phone’s back camera.

This subgraph only supports orientations of up to +/- 45°.

Input streams:

IMAGE: The ImageFrame stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_full_range_cpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_full_range_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_full_range_cpu_graph

Code: source code

FaceDetectionFullRangeSubgraphGpu

A full-range face detection subgraph.

This subgraph only supports full-range detection, e.g. from a phone’s back camera.

This subgraph only supports orientations of up to +/- 45°.

Input streams:

IMAGE: The GpuBuffer stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_full_range_gpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_full_range_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_full_range_gpu_graph

Code: source code

FaceDetectionRotatedFullRangeSubgraphGpu

A full-range face detection subgraph, with an extra input stream for the camera’s orientation.

This subgraph utilizes a face detection method that only supports orientations of up to +/- 45°. However, it uses the ROTATION_DEGREES input stream to identify the camera’s orientation. Thus the input image is rotated by that amount, the detection method is applied in the rotated image, and the resulting detections are rotated back.

This subgraph only supports full-range detection, e.g. from a phone’s back camera.

Input streams:

IMAGE: The GpuBuffer stream containing the image on which faces will be detected.
ROTATION_DEGREES: An int stream representing the camera’s orientation in degrees, counter-clockwise. Only 0, 90, 180, 270 are considered valid inputs.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_rotated_full_range_gpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_rotated_full_range_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_rotated_full_range_gpu_graph

Code: source code

FaceDetectionShortAndFullRangeSubgraphCpu

A face detection subraph that supports both short and full ranges.

This subgraph only supports orientations of up to +/- 45°.

This subgraph applies separate short and full-range detections, and then applies a non-max suppression to remove duplicate detections.

Input streams:

IMAGE: The ImageFrame stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_short_and_full_range_cpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_short_and_full_range_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_short_and_full_range_cpu_graph

Code: source code

FaceDetectionShortAndFullRangeSubgraphGpu

A face detection subraph that supports both short and full ranges.

This subgraph only supports orientations of up to +/- 45°.

This subgraph applies separate short and full-range detections, and then applies a non-max suppression to remove duplicate detections.

Input streams:

IMAGE: The GpuBuffer stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_short_and_full_range_gpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_short_and_full_range_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_short_and_full_range_gpu_graph

Code: source code

FaceDetectionShortRangeSubgraphCpu

A short-range face detection subgraph.

This subgraph only supports short-range detection, e.g. from a phone’s front camera.

This subgraph only supports orientations of up to +/- 45°.

Input streams:

IMAGE: The ImageFrame stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_short_range_cpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_short_range_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_short_range_cpu_graph

Code: source code

FaceDetectionShortRangeSubgraphGpu

A short-range face detection subgraph.

This subgraph only supports short-range detection, e.g. from a phone’s front camera.

This subgraph only supports orientations of up to +/- 45°.

Input streams:

IMAGE: The GpuBuffer stream containing the image on which faces will be detected.

Output streams:

DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/detection:face_detection_short_range_gpu

Text proto file:

@magritte//magritte/graphs/detection:face_detection_short_range_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/detection:face_detection_short_range_gpu_graph

Code: source code

Redaction

FaceDetectionToMaskSubgraphCpu

Converts face detections into a mask. The mask will have oval shapes at the location of each face, which are rotated by the angle given by the position of the eyes for each face detection. More precisely, this will take the bounding box for each face detection, slightly enlarge it, rotate it so that its edges are parallel and perpendicular to the line connecting the eyes, and inscribe an oval into the resulting rectangle. The mask background will be black and the ovals will be white.

Input streams:

IMAGE: An ImageFrame used to determine the size and format of the mask. The mask will have the same resolution as this, and use the same image format.
DETECTIONS: Face detections.

Output streams:

MASK: ImageFrame containing the created mask.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction/detection_to_mask:face_detection_to_mask_cpu

Text proto file:

@magritte//magritte/graphs/redaction/detection_to_mask:face_detection_to_mask_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction/detection_to_mask:face_detection_to_mask_cpu_graph

Code: source code

FaceDetectionToMaskSubgraphGpu

Input streams:

IMAGE: A GpuBuffer used to determine the size and format of the mask. The mask will have the same resolution as this, and use the same image format.
DETECTIONS: Face detections.

Output streams:

MASK: GpuBuffer containing the created mask.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction/detection_to_mask:face_detection_to_mask_gpu

Text proto file:

@magritte//magritte/graphs/redaction/detection_to_mask:face_detection_to_mask_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction/detection_to_mask:face_detection_to_mask_gpu_graph

Code: source code

DetectionTrackingOverlaySubgraphCpu

Subgraph to draw debug information at the locations specified by two incoming detection streams. For each detection stream, it will draw all data available from detections, including the bounding box, keypoints and score.

Input streams:

IMAGE: An ImageFrame containing the image to draw the overlays on.
[0]: A std::vector<mediapipe::Detection>, will be rendered in red.
[1]: A std::vector<mediapipe::Detection>, will be rendered in blue.

Output streams:

IMAGE: The resulting image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction:detection_tracking_overlay_cpu

Text proto file:

@magritte//magritte/graphs/redaction:detection_tracking_overlay_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction:detection_tracking_overlay_cpu_graph

Code: source code

DetectionTrackingOverlaySubgraphGpu

Input streams:

IMAGE_GPU: A GpuBuffer containing the image to draw the overlays on.
[0]: A std::vector<mediapipe::Detection>, will be rendered in red.
[1]: A std::vector<mediapipe::Detection>, will be rendered in blue.

Output streams:

IMAGE_GPU: The resulting image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction:detection_tracking_overlay_gpu

Text proto file:

@magritte//magritte/graphs/redaction:detection_tracking_overlay_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction:detection_tracking_overlay_gpu_graph

Code: source code

FaceDetectionOverlaySubgraphCpu

Subgraph to draw debug information at the locations specified by incoming detections. It will draw the detection bounding boxes as red un-filled rectangles with keypoints and detection score. In addition, it will draw green un-filled ovals around the area to be redacted.

Input streams:

IMAGE: An ImageFrame containing the image to draw the overlays on.
DETECTIONS: Face detections.

Output streams:

IMAGE: The resulting image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction:face_detection_overlay_cpu

Text proto file:

@magritte//magritte/graphs/redaction:face_detection_overlay_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction:face_detection_overlay_cpu_graph

Code: source code

FaceDetectionOverlaySubgraphGpu

Input streams:

IMAGE: A GpuBuffer containing the image to draw the overlays on.
DETECTIONS: Face detections.

Output streams:

IMAGE: The resulting image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction:face_detection_overlay_gpu

Text proto file:

@magritte//magritte/graphs/redaction:face_detection_overlay_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction:face_detection_overlay_gpu_graph

Code: source code

FacePixelizationSubgraphCpu

A subgraph that pixelizes faces.

This subgraph utilizes the mask pixelization: a mask is created based on the face detections, which is then used to blend the input with a pixelized version of the whole image.

Input streams:

IMAGE: An ImageFrame stream containing the image to be pixelized.
DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Output streams:

IMAGE: An ImageFrame stream containing the pixelized image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction:face_pixelization_cpu

Text proto file:

@magritte//magritte/graphs/redaction:face_pixelization_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction:face_pixelization_cpu_graph

Code: source code

FacePixelizationSubgraphGpu

A subgraph that pixelizes faces.

This subgraph utilizes a calculator that blends the pixelized image only in the redacted areas. This is about 3.7 times faster than using the mask pixelization, but less general-purpose.

Input streams:

IMAGE: A GpuBuffer stream containing the image to be pixelized.
DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Output streams:

IMAGE: A GpuBuffer stream containing the pixelized image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction:face_pixelization_gpu

Text proto file:

@magritte//magritte/graphs/redaction:face_pixelization_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction:face_pixelization_gpu_graph

Code: source code

FaceStickerRedactionSubgraphCpu

A subgraph that redacts faces with a custom “sticker” image.

The sticker is positioned and zoomed so that it covers the bounding box of each detected face. A default extra zoom can be applied via an input side packet, e.g. if the sticker has a transparent border.

Input streams:

IMAGE: The ImageFrame stream containing the image to be redacted.
DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Output streams:

IMAGE: An ImageFrame stream containing the redacted image.

Input side packets:

STICKER_PATH: The path for the sticker image file. If not provided, defaults to emoji.png.
STICKER_ZOOM: The default extra sticker zoom as a float. If not provided, the default value is 1.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction:face_sticker_redaction_cpu

Text proto file:

@magritte//magritte/graphs/redaction:face_sticker_redaction_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction:face_sticker_redaction_cpu_graph

Code: source code

FaceStickerRedactionSubgraphGpu

A subgraph that redacts faces with a custom “sticker” image.

Input streams:

IMAGE: The GpuBuffer stream containing the image to be redacted.
DETECTIONS: A list of face detections as std::vector<mediapipe::Detection>.

Output streams:

IMAGE: A GpuBuffer stream containing the redacted image.

Input side packets:

STICKER_PATH: The path for the sticker image file. If not provided, defaults to emoji.png.
STICKER_ZOOM: The default extra sticker zoom as a float. If not provided, the default value is 1.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction:face_sticker_redaction_gpu

Text proto file:

@magritte//magritte/graphs/redaction:face_sticker_redaction_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction:face_sticker_redaction_gpu_graph

Code: source code

MaskPixelizationSubgraphCpu

Subgraph to pixelize a certain area of an image. The mask specifies which area is to be pixelized: the parts where the mask is black will contain the original image, the parts where the mask is white will contain the pixelized image, and for values in-between blending will be used.

Input streams:

IMAGE: An ImageFrame containing the image to be pixelized.
MASK: An ImageFrame, containing a mask in ImageFormat::VEC32F1 format. It doesn’t need to have the same resolution as the input image, if it has a different resolution it will be scaled using cv::INTER_NEAREST interpolation.

Output streams:

IMAGE: An ImageFrame containing the resulting image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction/mask_redaction:mask_pixelization_cpu

Text proto file:

@magritte//magritte/graphs/redaction/mask_redaction:mask_pixelization_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction/mask_redaction:mask_pixelization_cpu_graph

Code: source code

MaskPixelizationSubgraphGpu

Input streams:

IMAGE: A GpuBuffer containing the image to be pixelized.
MASK: A GpuBuffer containing a mask. The image format of the mask doesn’t matter, we always use the first channel as the alpha value for blending. It doesn’t need to have the same resolution as the input image, if it has a different resolution it will be scaled using linear interpolation (in most cases, see SetStandardTextureParams in MediaPipe’s GlContext class for details).

Output streams:

IMAGE: A GpuBuffer containing the resulting image.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/redaction/mask_redaction:mask_pixelization_gpu

Text proto file:

@magritte//magritte/graphs/redaction/mask_redaction:mask_pixelization_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/redaction/mask_redaction:mask_pixelization_gpu_graph

Code: source code

Tracking

SampledTrackingSubgraphCpu

A graph that performs motion tracking on detections, and in addition performs sampling on the input video stream. It is similar to the tracking graphs without sampling in this directory, and in addition it outputs a downsampled version of the input video stream with 5 fps. This downsampled video stream can be used for performing detections that are fed back into this graph.

Input streams:

input_video: The ImageFrame stream in which objects should be tracked. Its motion will be analyzed for the tracking.
sampled_detections: The detections to be tracked. They are expected to be calculated from the sampled_input_video output stream of this graph.

Output streams:

tracked_detections: Resulting tracked detections.
sampled_input_video: The input video stream downsampled to 5 fps.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/tracking:sampled_tracking_cpu

Text proto file:

@magritte//magritte/graphs/tracking:sampled_tracking_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/tracking:sampled_tracking_cpu_graph

Code: source code

SampledTrackingSubgraphGpu

Input streams:

input_video: The GpuBuffer stream in which objects should be tracked. Its motion will be analyzed for the tracking.
sampled_detections: The detections to be tracked. They are expected to be calculated from the sampled_input_video output stream of this graph.

Output streams:

tracked_detections: Resulting tracked detections.
sampled_input_video: The input video stream downsampled to 5 fps.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/tracking:sampled_tracking_gpu

Text proto file:

@magritte//magritte/graphs/tracking:sampled_tracking_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/tracking:sampled_tracking_gpu_graph

Code: source code

TrackingSubgraphCpu

A graph that performs motion tracking on detections. It’s a wrapper around the MediaPipe object tracking subgraph.

The input streams don’t need to have the same timestamps. The detections stream can be at a lower rate (e.g., obtained from a video that was downsampled to a lower framerate). The output detection stream will be generated for the same timestamps as the input video packets.

Input streams:

input_video: The ImageFrame stream in which objects should be tracked. Its motion will be analyzed for the tracking.
detections: The detections to be tracked.

Output streams:

tracked_detections: Resulting tracked detections.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/tracking:tracking_cpu

Text proto file:

@magritte//magritte/graphs/tracking:tracking_cpu.pbtxt

Binary graph:

@magritte//magritte/graphs/tracking:tracking_cpu_graph

Code: source code

TrackingSubgraphGpu

A graph that performs motion tracking on detections. It’s a wrapper around the MediaPipe object tracking subgraph.

Input streams:

input_video: The GpuBuffer stream in which objects should be tracked. Its motion will be analyzed for the tracking.
detections: The detections to be tracked.

Output streams:

tracked_detections: Resulting tracked detections.

Build targets:

Graph cc_library:

@magritte//magritte/graphs/tracking:tracking_gpu

Text proto file:

@magritte//magritte/graphs/tracking:tracking_gpu.pbtxt

Binary graph:

@magritte//magritte/graphs/tracking:tracking_gpu_graph

Code: source code