YOLO label recovery tool that audits and auto-fixes missing annotations
The project provides a comprehensive pipeline to audit YOLO datasets for missing or incorrect labels and automatically recover them using multiple class‑specific teachers. It offers model‑free checks, threshold calibration, consensus gating, near‑duplicate grouping, and active review prioritization, all without modifying the original dataset. It is aimed at data engineers and ML practitioners who need reliable, auditable label correction for multi‑class vision tasks, delivering a more robust dataset than ad‑hoc labeling scripts.
View on GitHub →jiapengLi11/yolo-label-recovery