MediaPipeTasksVision Framework Reference

ImageEmbedderOptions

class ImageEmbedderOptions : TaskOptions, NSCopying

Options for setting up a ImageEmbedder.

  • Running mode of the image embedder task. Defaults to .image. ImageEmbedder can be created with one of the following running modes:

    1. .image: The mode for performing embedding extraction on single image inputs.
    2. .video: The mode for performing embedding extraction on the decoded frames of a video.
    3. .liveStream: The mode for performing embedding extraction on a live stream of input data, such as from the camera.

    Declaration

    Swift

    var runningMode: RunningMode { get set }

  • An object that confirms to ImageEmbedderLiveStreamDelegate protocol. This object must implement imageEmbedder(_:didFinishEmbeddingWithResult:timestampInMilliseconds:error:) to receive the results of asynchronous embedding extraction on images (i.e, when runningMode = .liveStream).

    Declaration

    Swift

    weak var imageEmbedderLiveStreamDelegate: ImageEmbedderLiveStreamDelegate? { get set }

  • @brief Sets whether L2 normalization should be performed on the returned embeddings. Use this option only if the model does not already contain a native L2_NORMALIZATION TF Lite Op. In most cases, this is already the case and L2 norm is thus achieved through TF Lite inference.

    NO by default.

    Declaration

    Swift

    var l2Normalize: Bool { get set }

  • @brief Sets whether the returned embedding should be quantized to bytes via scalar quantization. Embeddings are implicitly assumed to be unit-norm and therefore any dimensions is guaranteed to have value in [-1.0, 1.0]. Use the l2Normalize property if this is not the case.

    NO by default.

    Declaration

    Swift

    var quantize: Bool { get set }