Skip to main content

Module synthetic

Module synthetic 

Source
Expand description

Phase 7 / V7 — regenerable synthetic labeled scenes (no Python, no network).

Deterministic from seed: colored rectangles on a dark field with ground-truth boxes. Train/test split uses disjoint seed ranges so labels cannot leak.

Structs§

SyntheticSampleId
Manifest entry for one regenerable sample.

Enums§

DatasetSplit
Dataset partition — train and test use disjoint seed bases.

Constants§

CLASS_MOSTLY_BLUE
CLASS_MOSTLY_GREEN
CLASS_MOSTLY_RED
TEST_SEED_BASE
TRAIN_SEED_BASE
Train seeds: TRAIN_SEED_BASE + i. Test: TEST_SEED_BASE + i. Never overlap.

Functions§

generate_scene_rgb8
Generate RGB8 scene into out (len ≥ wh3). Returns ground-truth count written to gt.
match_accuracy
Simple IoU match accuracy: fraction of GT boxes that have a prediction with IoU ≥ thresh and same class. Used for synthetic uplift checks (not a full COCO evaluator).
sample_id
Build a sample id. index is 0-based within the split.
train_test_disjoint
Assert train/test seed bases never collide for the first n indices of each.