xellDart

Plan: Modulo de Video (tipo YouTube) para Big2Be API

Contexto

Big2Be es una plataforma de eventos y experiencias. Se necesita agregar un modulo de video completo estilo YouTube para que los creadores de contenido puedan subir videos y shorts relacionados con eventos, behind-the-scenes, promociones, y experiencias. El modulo incluye: upload, procesamiento con FFmpeg, streaming HLS adaptativo, comentarios, likes, shares, tracking de vistas, y moderacion de contenido con IA via OpenRouter.

Decisiones del usuario:

• Storage: AWS S3 (ya lo tienen configurado)
• Moderacion: OpenRouter (gateway AI multi-modelo con vision)
• Tipo de plataforma: Eventos y experiencias (Big2Be focused)

xellDart

List<Detection> origNms(List<Detection> detections, double threshold) {
    if (detections.length <= 0) return [];
    List<double> x1 = new List();
    List<double> x2 = new List();
    List<double> y1 = new List();
    List<double> y2 = new List();
    List<double> s = new List();

    detections.forEach((detection) {
      x1.add(detection.xMin);

xellDart

Array decodeBox(
      List<double> rawBoxes, int i, List<Anchor> anchors, OptionsFace options) {
    Array boxData = Array(List<double>.generate(options.numCoords, (i) => 0.0));
    int boxOffset = i * options.numCoords + options.boxCoordOffset;
    double yCenter = rawBoxes[boxOffset];
    double xCenter = rawBoxes[boxOffset + 1];
    double h = rawBoxes[boxOffset + 2];
    double w = rawBoxes[boxOffset + 3];
    if (options.reverseOutputOrder) {
      xCenter = rawBoxes[boxOffset];

xellDart

Detection convertToDetection(double boxYMin, double boxXMin, double boxYMax,
      double boxXMax, double score, int classID, bool flipVertically) {
    double _yMin;
    if (flipVertically)
      _yMin = 1.0 - boxYMax;
    else
      _yMin = boxYMin;
    return new Detection(score, classID, boxXMin, _yMin, (boxXMax - boxXMin),
        (boxXMax - boxYMin));
  }

xellDart

List<Anchor> getAnchors(AnchorOption options) {
    List<Anchor> _anchors = new List();
    if (options.stridesSize != options.numLayers) {
      print('strides_size and num_layers must be equal.');
      return [];
    }
    int layerID = 0;
    while (layerID < options.stridesSize) {
      List<double> anchorHeight = new List();
      List<double> anchorWidth = new List();

xellDart

 List<Detection> process(
      {OptionsFace options,
      List<double> rawScores,
      List<double> rawBoxes,
      List<Anchor> anchors}) {
    List<double> detectionScores = new List();
    List<int> detectionClasses = new List();

    int boxes = options.numBoxes;
    for (int i = 0; i < boxes; i++) {

xellDart

// Load model
void _loadModel() async {
    _anchors = UtilsFace().getAnchors(anchors);
    _interpreter =
        await Interpreter.fromAsset("models/face_detection_front.tflite");
    _inputShape = _interpreter.getInputTensor(0).shape;
    _imageProcessor = ImageProcessorBuilder()
        .add(ResizeOp(
            _inputShape[1], _inputShape[2], ResizeMethod.NEAREST_NEIGHBOUR))
        .add(_normalizeInput)

xellDart

class Detection {
  final double score;
  final int classID;
  final double xMin;
  final double yMin;
  final double width;
  final double height;
  Detection(
      this.score, this.classID, this.xMin, this.yMin, this.width, this.height);
}

xellDart

class AnchorOption {
  final int inputSizeWidth;
  final int inputSizeHeight;
  final double minScale;
  final double maxScale;
  final double anchorOffsetX;
  final double anchorOffsetY;
  final int numLayers;
  final List<int> featureMapWidth;
  final List<int> featureMapHeight;

xellDart

class OptionsFace {
  final int numClasses;
  final int numBoxes;
  final int numCoords;
  final int keypointCoordOffset;
  final List<int> ignoreClasses;
  final double scoreClippingThresh;
  final double minScoreThresh;
  final int numKeypoints;
  final int numValuesPerKeypoint;