YOLO Label Recovery v0.6.0

English

YOLO Label Recovery v0.6.0 adds a model-free cross-Teacher consensus gate for safer automatic label recovery.

• Independent primary/verifier candidate streams with no simultaneous model residency
• Deterministic one-to-one spatial matching by image, class and IoU
• Unsupported primary AUTO candidates downgrade to REVIEW instead of being discarded
• Optional agreed-only YOLO label additions
• JSON, CSV and self-contained HTML audit evidence
• Reproducible six-class fixture and bilingual portfolio documentation
• 28 local automated checks plus GitHub Actions CI

??

YOLO Label Recovery v0.6.0 ???????? Teacher ?????,?????????

• ? Teacher ??? Teacher ????,?????????????
• ??????? IoU ????????????
• ?????? AUTO ??? REVIEW,???????
• ?????????????? YOLO ????
• ?? JSON?CSV ???? HTML ????
• ??????????????????
• ?? 28 ?????,?? GitHub Actions ????

YOLO Label Recovery v0.5.0

English

YOLO Label Recovery v0.5.0 upgrades class-specific AUTO threshold calibration from raw empirical precision to a Wilson score precision lower bound.

• 95% confidence-aware AUTO gating by default
• Empirical compatibility mode with --auto-confidence-level 0
• 2,400-candidate, six-class reproducible public fixture
• JSON, CSV and HTML evidence for empirical precision and precision LCB
• 23 local automated checks plus GitHub Actions CI

??

YOLO Label Recovery v0.5.0 ???? AUTO ?????????? Wilson ???????

• ???? 95% ????????????
• ??? --auto-confidence-level 0 ????????
• ??????2,400 ???????????
• JSON?CSV?HTML ???????????????
• ?? 23 ?????,?? GitHub Actions ????

YOLO Label Recovery v0.4.0

English

• Added audited, class-specific AUTO precision and REVIEW recall calibration.
• Added JSON policy output, threshold-curve CSV and a self-contained HTML report.
• Added a deterministic six-class, 600-candidate public fixture and static portfolio screenshot.
• Expanded automated coverage to 21 checks across smoke and pytest suites.

中文

• 新增基于人工审核结果的分类别 AUTO 精度与 REVIEW 召回阈值校准。
• 新增 JSON 策略、阈值曲线 CSV 和自包含 HTML 报告。
• 新增六类别、600 条候选的可复现公开样本和静态面试展示截图。
• 冒烟测试与 pytest 合计扩展到 21 项检查。

YOLO Label Recovery v0.3.1

English

• Added complete Simplified Chinese documentation and language navigation.
• Split lightweight audit/report installation from optional GPU inference dependencies.
• Added actionable optional-dependency diagnostics and test coverage.

中文

• 新增完整简体中文首页、语言切换和中文面试讲解稿。
• 将轻量审计/报告依赖与 GPU 推理依赖拆分，公开演示安装更快。
• 增加推理依赖缺失提示及对应自动化测试。

YOLO Label Recovery v0.3.0

First public release of the safe multi-teacher YOLO missing-label recovery pipeline.

Highlights:

• model-free dataset audit and split-leakage detection
• sequential teacher inference with adaptive OOM recovery
• atomic checkpoints and idempotent resume
• runtime doctor, reproducibility manifests and HTML reports
• synthetic public demo, tests and GitHub Actions CI