Enciclopedia Boulder v1.0

Inicio · Capítulos · 14. Señales para visión artificial

15.202 14.7.28 Pipeline completo del dataset

Etapa ID Acción
1 DS-01-GOVERN Define intended use, prohibited use, risk tier, consent and release levels.
2 DS-02-DESIGN Freeze ontology/guideline draft, sampling plan, units, metrics and split constraints.
3 DS-03-COLLECT Capture multimodal sessions with device, scene, clock and governance metadata.
4 DS-04-INGEST Hash, validate, quarantine corruption and create immutable raw manifests.
5 DS-05-PREPROCESS Generate deterministic proxies, redactions, clips and alignment products.
6 DS-06-UNITIZE Create attempts and annotation units without leakage across future groups.
7 DS-07-PILOT Run representative pilot and revise ontology/guidelines before scale.
8 DS-08-QUALIFY Train and calibrate annotators using examples and blind qualification tasks.
9 DS-09-ANNOTATE Independent layered annotation with uncertainty and evidence ledger.
10 DS-10-AUTO-QA Schema, referential, temporal, geometric and physical checks.
11 DS-11-CONSENSUS Compute agreement and candidates without deleting individual judgments.
12 DS-12-REVIEW Manual review, defect coding and correction as new revisions.
13 DS-13-ADJUDICATE Expert resolution or explicit ambiguity; trigger ontology issues.
14 DS-14-AUDIT Blind gold, spot checks, annotator drift and batch-level error budgets.
15 DS-15-COVERAGE Assess long tail, unknowns, contrast sets, domains and privacy-safe cohorts.
16 DS-16-DEDUP Exact/perceptual duplicate detection before splitting.
17 DS-17-SPLIT Immutable grouped splits and leakage audit.
18 DS-18-PACKAGE Canonical data, adapters, Croissant, PROV, cards and licenses.
19 DS-19-GATE Execute quality, privacy, security and reproducibility gates.
20 DS-20-RELEASE Sign release, publish allowed tier and archive root hashes.
21 DS-21-MONITOR Track downstream issues, drift, misuse and withdrawal requests.
22 DS-22-ITERATE Use active learning only on eligible pools; create new version and preserve lineage.