Crate visual_cortex_vision

Expand description

Detectors: the Detector trait, built-in gates, and the OcrEngine trait.

Modules§

filters: Stock span filters for OCR detectors. A filter returns true to KEEP a span (like Iterator::filter); filtered spans are removed BEFORE change detection, so noise (a flickering button glyph) cannot fire spurious Changed events.
patterns: Text parsers for OCR output. Used with WatcherBuilder::ocr.

Brightness: Mean luma (0.0..=255.0) of the region, as DetectorOutput::Number. Pair with .on_threshold(...) for “screen went dark”-style watchers.
ColorDominance: Boolean detector/gate: at least min_fraction of pixels are dominated by one channel (e.g. “the HP bar area is mostly red”).
ColorMatch: Boolean detector/gate: at least min_fraction of the region’s pixels are within tolerance (per channel) of the target RGB color.
FnDetector: Adapt a closure into a Detector. See from_fn.
FrameDiff: Boolean gate: did the region’s pixels change since the last evaluation? Hashes the region’s bytes; the first evaluation always reports changed.
OcrDetector
PngDump: PNG stills per sampled tick: <dir>/<watcher>/NNNNNN-{input,baseline,overlay,output}.png.
StabilityMask: Temporal stability mask: passes through only newly-appeared, now-static overlays (item tooltips, dialogs); moving gameplay and the permanent scene render black. Kept blocks are byte-identical to the input. Tick it fast (5–10 Hz): the frame-diff gate downstream keeps detector cost unchanged while nothing changes.
TemplateMatcher: Grayscale normalized cross-correlation template matcher.
TextSpan: One recognized text region. bbox is relative to the evaluated view.

DebugSink: A tap receiving per-tick stage frames from a preprocessor (see StabilityMask::debug_sink).
Detector: Evaluates one region of one frame. Built-in detectors, gates, and user-defined detectors all implement this same trait.
OcrEngine: Pluggable OCR backend. The default ONNX/PaddleOCR implementation lives in the visual-cortex-ocr-onnx crate; any engine can slot in here. Returning an empty vec means “no text present” and is not an error.
Preprocessor: Transforms a watcher’s cropped view into the frame the rest of the pipeline sees. Runs BEFORE the implicit frame-diff gate, so a preprocessor whose output is static (e.g. a fully-suppressed stability mask) holds the detector at zero cost until its output actually changes.

from_fn: Build a detector from a closure: from_fn(|view| Ok(DetectorOutput::Bool(true))).