MpFaceTrackingMultiElement Op

read more on the mediapipe project page: https://chuoling.github.io/mediapipe/solutions/face_mesh.html

some info about the mesh data: https://github.com/lysdexic-audio/n4m-facemesh

Overview of Improvements markdown_content = """# FaceMesh Operator: Multi-Face Tracking Documentation Update

Overview of Improvements

The upgraded operator transitions from a single-face tracking prototype to a production-ready, multi-face tracking node with improved frame scheduling, coordinate stabilization, and resource cleanup.

• Simultaneous Multi-Face Tracking: Tracks up to $N$ faces simultaneously instead of hardcoding detection to a single face.
• Spatial Index Sorting: Eliminates index-swapping jitter across frames by sorting detected faces along the X-axis.
• Bounding Box Generation: Computes normalized 2D bounding boxes (center X/Y, width, height) in real time for each detected face.
• Custom Render Loop & State Protection: Replaces the third-party Camera utility with an asynchronous requestAnimationFrame loop featuring concurrency gating (isProcessing) and video readyState checks to prevent MediaPipe freezes.
• Lifecycle & Memory Management: Implements op.onDelete to cancel active animation frames and call faceMesh.close(), preventing WebAssembly (WASM) memory leaks on operator recompilation or deletion.

New Input Parameters

• Max Faces (inMaxFaces | Integer, default: 4): Defines the maximum number of faces MediaPipe should detect and track simultaneously.
• Min Detection Confidence (inMinConfidence | Float Slider, default: 0.5): Sets the minimum confidence score ($[0.0, 1.0]$) for the face detection model to consider a detection successful.
• Min Tracking Confidence (inMinTracking | Float Slider, default: 0.5): Sets the minimum confidence score ($[0.0, 1.0]$) for landmark tracking; higher values reduce jitter but drop fast-moving faces.
• Sort Faces (L to R) (inSortFaces | Boolean, default: true): Orders output arrays from left to right based on face center X coordinates, preventing target index swapping when people shift positions.
• Flip X (inFlipX | Boolean, default: false): Inverts coordinates along the horizontal axis, providing mirrored mapping suitable for standard selfie/webcam feeds.

Output Changes

• Points (outPoints | Array): Now outputs an array of point arrays (allPoints[faceIndex]), where each nested array contains the normalized 3D coordinates $(x, y, 0)$ for that face.
• Bounding Boxes (outBoxes | Array): Emits an array of 4D bounding boxes formatted as [centerX, centerY, boxWidth, boxHeight] for each detected face.
• Found Faces (outFound | Number): Outputs the total number of faces successfully identified in the current frame.

Multiface tracking Op with bounding box and videoTextureElement Op example

• Inspired by https://cables.gl/p/fr5hEj and https://cables.gl/p/Q78AyQ

• New features added (check Op's documentation)