VisionIntent

enum class MayaFlux::Kinesis::Vision::VisionIntent : uint32_t

strong

Composable flags naming what a vision analysis should produce.

Backed by today's VisionOp catalog: FindElements, TrackObjects, DetectFeatures, DetectEdges, EstimateMotion, MeasureAppearance.

Named but not yet backed by any VisionOp; resolving one of these is expected to fail clearly until a supporting op exists, the same way VisionGpuExecutor::config() returns INVALID_SHADER for an op with no GPU implementation rather than omitting the op: Classify, EstimateDepth, DetectAnomaly, Register.

Enumerator
NONE
FindElements	Localize discrete things in the frame (thresholding/morphology into ConnectedComponents/FindContours).
TrackObjects	Maintain identity of specific points across frames (ExtractPeaks into TrackKeypoints).
DetectFeatures	Distinctive points with no persistence claim (HarrisResponse into ExtractPeaks alone).
DetectEdges	Boundary/gradient detection (Sobel, Scharr, Canny).
EstimateMotion	Dense, whole-field motion with no identity claim (OpticalFlowDense). Kept separate from TrackObjects: a global flow field and a persistently tracked point are different claims.
MeasureAppearance	Photometric/statistical characterization of the frame itself, independent of anything found in it.
Classify	Assign a label to a detected region or the whole frame. No backing VisionOp exists yet; needs a learned-model op.
EstimateDepth	Per-pixel depth or disparity. No backing VisionOp exists yet; a different output shape entirely from every flag above it.
DetectAnomaly	Flag departure from a learned or authored baseline. No backing VisionOp exists yet; needs a reference/baseline concept VisionOp has no notion of today.
Register	Align two or more frames into a common reference. No backing VisionOp exists yet; the only flag here whose input is plural.

Definition at line 36 of file VisionIntent.hpp.

                        : uint32_t {
    NONE = 0U,
 
    /// Localize discrete things in the frame (thresholding/morphology into
    /// ConnectedComponents/FindContours).
    FindElements = 1U << 0U,
 
    /// Maintain identity of specific points across frames (ExtractPeaks
    /// into TrackKeypoints).
    TrackObjects = 1U << 1U,
 
    /// Distinctive points with no persistence claim (HarrisResponse into
    /// ExtractPeaks alone).
    DetectFeatures = 1U << 2U,
 
    /// Boundary/gradient detection (Sobel, Scharr, Canny).
    DetectEdges = 1U << 3U,
 
    /// Dense, whole-field motion with no identity claim (OpticalFlowDense).
    /// Kept separate from TrackObjects: a global flow field and a
    /// persistently tracked point are different claims.
    EstimateMotion = 1U << 4U,
 
    /// Photometric/statistical characterization of the frame itself,
    /// independent of anything found in it.
    MeasureAppearance = 1U << 5U,
 
    /// Assign a label to a detected region or the whole frame. No backing
    /// VisionOp exists yet; needs a learned-model op.
    Classify = 1U << 6U,
 
    /// Per-pixel depth or disparity. No backing VisionOp exists yet; a
    /// different output shape entirely from every flag above it.
    EstimateDepth = 1U << 7U,
 
    /// Flag departure from a learned or authored baseline. No backing
    /// VisionOp exists yet; needs a reference/baseline concept VisionOp
    /// has no notion of today.
    DetectAnomaly = 1U << 8U,
 
    /// Align two or more frames into a common reference. No backing
    /// VisionOp exists yet; the only flag here whose input is plural.
    Register = 1U << 9U,
};