Category description and use cases
To allow collection managers to locate known persons in collections materials. If, for example, a collection has many images of someone important to their institution and suspects they appear in video footage but would like to confirm, or would like to know where in a video the person appears, a face recognition tool would provide that information.
Wrapper for metadata about the source media file.
Filename of the source file.
The duration of the source file.
The frame rate of the video, in FPS.
The number of frames in the video.
Resolution of the video.
Width of the frame, in pixels.
Height of the frame, in pixels.
List of frames containing identified faces.
A frame containing an identified face.
Time of the frame, in seconds.
List of bounding boxes in the frame containing identified faces.
A bounding box in the frame containing an identified face.
The name of the face within the bounding box.
A confidence or relevance score for the face.
string (confidence | relevance)
The type of score, confidence or relevance.
The score value, typically a number in the range of 0-1.
The top left (xmin, ymin) and bottom right (xmax, ymax) relative bounding coordinates.
The top left x coordinate.
The top left y coordinate.
The bottom right x coordinate.
The bottom right y coordinate.
Description: OpenCV-based face recognition library.
Cost: Free (open source)
Social impact: We retain full control of use of the images/face data.
Notes: Tests run on Charlie Nelms and Herman B Wells images/videos.
Installation & requirements
Install via pip (face_recognition).
For training: Images labelled with person's name (currently via file path, but this should perhaps change-- discussion to have with dev)
For identifying: A model trained on the relevant people
See Colab notebook.
List of timestamps where face was found