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§
- Synthetic
Sample Id - Manifest entry for one regenerable sample.
Enums§
- Dataset
Split - 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 togt. - 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.
indexis 0-based within the split. - train_
test_ disjoint - Assert train/test seed bases never collide for the first
nindices of each.