exordium.video.face.streaming

Live face tracking: frames in, per-frame tracks and a provisional protagonist out.

Attributes

logger

Module-level logger.

Classes

StreamingFacePipeline

Detect and track faces on frames as they arrive.

Module Contents

exordium.video.face.streaming.logger

Module-level logger.

class exordium.video.face.streaming.StreamingFacePipeline(detector: exordium.video.face.detector.base.FaceDetector, tracker: exordium.video.core.streaming.StreamingTracker)

Detect and track faces on frames as they arrive.

Chains a FaceDetector with a StreamingTracker; nothing is read ahead, so it works on a live camera, a call, or any frame iterable.

Parameters:

  • detector – Face detector run on every frame.

  • tracker – Streaming tracker fed with the detections.

Example:

pipeline = StreamingFacePipeline(
    YoloFace11Detector(), StreamingTracker(IouTracker())
)
for update in pipeline.process(frames):       # frames: (3, H, W) uint8 RGB
    if update.stable:
        print("protagonist:", update.protagonist_id)
push(frame: torch.Tensor | numpy.ndarray, frame_id: int | None = None) → exordium.video.core.streaming.StreamUpdate

Detect and track faces in the next frame.

Parameters:

  • frame – (3, H, W) tensor or (H, W, 3) array, uint8 RGB.

  • frame_id – Frame index; counts up from 0 when omitted.

Returns:

The StreamUpdate for this frame.

process(frames: collections.abc.Iterable[torch.Tensor | numpy.ndarray]) → collections.abc.Iterator[exordium.video.core.streaming.StreamUpdate]

Lazily yield one update per incoming frame.

Parameters:

frames – Any iterable of frames, such as a camera generator.

Yields:

One StreamUpdate per frame.

reset() → None

Forget all tracks and restart frame numbering for a new session.