Measure Image Sharpness (OpenCV)

How it fits

Typical backends

Caveats

• AccuracyGradient sharpness is content- and contrast-dependent: a texture-poor but in-focus region can score low while a high-contrast blurry one scores high, so tune per image class.
• Parameter interactionThe cutoff only transfers across cameras of differing pixel pitch when canonical_size rescales the longer edge first; left at zero, the raw input resolution is used and the cutoff must be re-tuned per source.
• Parameter interactionEach method produces scores on its own scale, and changing it requires recalibrating threshold; the gradient-variance options give the cleanest sharp/blur separation on text.
• I/O contractThe output arm is fixed at graph-build time from the downstream connection; a downstream type matching none of the supported shapes fails at startup, and an empty frame scores zero.