脚本语言
计算机科学
协议(科学)
数据挖掘
管道(软件)
绝缘体(电)
泄漏(经济)
审计
实时计算
软件
表(数据库)
人工智能
计算机硬件
在飞行中
数据采集
自动化
数据库
嵌入式系统
可靠性工程
计算机视觉
标识
DOI:10.5281/zenodo.22970563
摘要
This package accompanies the paper "Evaluation Bias in Vision-Based Insulator Defect Detection Systems: Quantification of Data Leakage, Impact, and a Leakage-Free Measurement Protocol". It contains (1) the three-level provenance audit pipeline (scene-ID screening, pixel-level twin verification, original-split comparison) that quantifies cross-split scene-level data leakage in the IDID and IFDD benchmarks; (2) the leakage-free evaluation subset idid_val_clean (186 images / 53 scenes, zero same-source overlap with the training set) and the scene-level re-split of IFDD (train 1456 / val 144) as manifest files; (3) 19 reproduction scripts covering leakage quantification, leakage-ratio response experiments, cross-architecture comparison, the random-label control, and third-domain (IFDD) protocol validation; (4) result snapshots backing every table of the paper. Original images are not redistributed; acquisition links for all datasets are provided in the README.
科研通智能强力驱动
Strongly Powered by AbleSci AI