ops

sleap_nn.inference.ops

Pure-ops library for inference (PR 1 of #508).

Stateless tensor operations grouped by concern. Every function here is:

• a pure function (no module state, no I/O)
• safe to import without pulling in Lightning, sleap-io, or video-reading deps
• the single source of truth — Lightning modules and (later) InferenceLayer subclasses both call into this namespace, so they can never drift

Sub-modules:

• :mod:peaks — global / local peak finding, integral refinement, morphological NMS
• :mod:paf — PAF scoring + grouping (PAFScorer)
• :mod:coord — coordinate-ladder reversers (stride / scale / crop)
• :mod:filters — keypoint and instance-level post-inference filters
• :mod:identity — multi-class peak grouping
• :mod:crops — bbox creation + cropping helpers

The original modules (peak_finding.py, paf_grouping.py etc.) are preserved as thin re-export shims so existing callers keep working through the deprecation window.

Modules:

Name	Description
coord	Coordinate-ladder reversers for inference.
crops	Bbox creation + image cropping helpers used by inference.
filters	Inference-level postprocessing filters for pose predictions.
identity	Utilities for models that learn identity.
paf	This module provides a set of utilities for grouping peaks based on PAFs.
peaks	Peak finding ops for confidence maps.