sidescanpy.export#

Dataset-export primitives for SideScan sonar surveys.

Distinct from processing/ (physical corrections: TVG, geometric ranges, georeferencing) and export/ml/ (ML annotation file formats): this package turns an already-corrected survey channel into 8-bit training-dataset images — chips or full-channel views, optionally reprojected into geometric or georeferenced meters — plus the shared invertible display transform they’re rendered through.

Public entry points#

to_uint8() / to_uint16() / from_uint8() — invertible percentile+gamma display transform, with a configurable per-image / precomputed (survey- or dataset-wide) normalization scope.
compute_reference_stats() — precompute percentile bounds shared across many to_uint8() calls, for survey-/dataset-wide scope.
sidescanpy.export.windows — where an image is cut and what is in it: extract_chip() / extract_full_channel() (plain arrays), extract_pixel_style_window() / extract_geo_style_window() (one export style at one render level), neighborhood loading, and sample_background_windows() (screened no-target windows).
sidescanpy.export.orientation — display orientation of exported windows and their targets.
sidescanpy.export.augment — random flips and jitter.
sidescanpy.export.ml — COCO/YOLO annotation writers.

2026 rshomberg@uri.edu

Modules