Expand description
Region learning over a whole recording: the core’s per-frame evidence accumulators
(crate::kernels::layers::Learner), the last informative native frame kept for appearance evidence, and the
one-shot region construction at the end — the automatic layout (crate::kernels::regions::finish) or a
hand-placed one (manual regions, plus the “unassigned” region when they leave pixels uncovered).
The scan session owns one of these; the web page’s LayerLearner (src/core/layers.ts) drives one directly
through rust/wasm/src/layers.rs for the layer-learning unit and parity tests.
Structs§
- Manual
Region - One hand-placed region from the run’s settings.
- Region
Learner - Layer-learning state for one run: accumulators plus the last informative native frame.
Constants§
- UNASSIGNED_
NAME - Name of the region that owns whatever native pixels a hand-placed layout leaves uncovered. Region names are a fixed zh vocabulary, translated only for display (src/i18n/names.ts).
Functions§
- label_
atlas - The native-resolution label plane for
regions(0 = no region, otherwise index + 1; the first region that contains a pixel owns it) and the per-code pixel counts, orNonepast 254 regions. Regions are described the way the page’s region rows carry them: an automatic region always has a crop (its own rect when it has none), and a solid region’s mask is not consulted. - manual_
uncovered - Whether some native pixel of a
native_width × native_heightframe lies in none ofrects. An empty list covers nothing.