--- title: verified-negatives-canaputxa canonical_url: "https://www.modelscope.ai/datasets/NKODATALABS/verified-negatives-canaputxa" md_url: "https://www.modelscope.ai/datasets/NKODATALABS/verified-negatives-canaputxa.md" repository: NKODATALABS/verified-negatives-canaputxa chinese_name: "已验证负样本-CaNaPutxa对照集" last_updated: 2026-09-05 license: cc-by-4.0 storage_size: "2.7 MB" downloads: 1 stars: 0 --- # verified-negatives-canaputxa > verified-negatives-canaputxa - An open-source dataset by NKODATALABS on ModelScope. 50 LiDAR-verified stable-ground polygons (verified negatives) from two independent PNOA-LiDAR campaigns (2019/2024), Ibiza, Spain. Hard negatives for change detection. CC-BY… NKODATALABS/verified-negatives-canaputxa is a dataset on ModelScope. totalling 2.7 MB. licensed under cc-by-4.0. - **Repository**: NKODATALABS/verified-negatives-canaputxa - **License**: cc-by-4.0 - **Storage size**: 2.7 MB - **Downloads**: 1 - **Stars**: 0 - **Last updated**: 2026-09-05 Source: https://www.modelscope.ai/datasets/NKODATALABS/verified-negatives-canaputxa --- # NKO Verified Negatives — Ca na Putxa control set (v0.1) *内部备注：本卡片由NKO直接翻译为简体中文并发布，翻译日期 2026-09-02，未经专业人工译者审校。* ## English 50 LiDAR-verified stable-ground polygons (verified negatives) from two independent PNOA-LiDAR campaigns (2019/2024), Ibiza, Spain. Hard negatives for change detection. CC-BY 4.0. A **hard negatives dataset** / **change detection ground truth**: 50 real, stable-terrain polygons — **LiDAR verified no-change** examples, each with `verdict=NO_CHANGE_DETECTED` between two real PNOA-LiDAR campaigns (2019-11-25 → 2024-04-16, dates confirmed via GpsTime embedded in the LiDAR points, not nominal filenames), produced by the same physical method already used to measure the Ca na Putxa landfill (Eivissa, Spain). ### Files - `polygons.geojson` — the 50 polygons with all properties, ready to load in QGIS/kepler.gl/Overpass turbo. - `polygons.parquet` — same table, columnar format. - `evidence/CTRL-*.json` — one evidence file per published polygon (50, exact 1:1 match with the table). - `passed_polygons.json` — the same 50 selected, flat JSON (alternate format of the same table). - `rejected_polygons.json` — the 4,327 real candidates that did NOT pass the coverage/threshold filter, complete, with rejection reason. - `METHOD.md` — exact methodology, with the flight-date limitation disclosed, not hidden. **Honest correction (2026-09-01):** the original script evaluated 4,757 real candidates that did pass the filter before this spatially-dispersed sample of 50 was drawn — but the full list of those 4,757 only existed in memory during the run and was never saved to disk. What wasn't kept can't be published: here are the 50 selected (complete, with evidence) and the 4,327 rejected (complete) — the 4,757-passed count is cited as a real figure from that run, not as a downloadable file. ### Why this matters Anyone can claim "I measured a real volume change." Very few also publish where their own method, applied to the same data, honestly says **"no signal here"** — same threshold, same minimum coverage, and the cases that didn't pass the filter listed separately, not hidden. That's what makes the Ca na Putxa number (+310,357 ± 183 m³, canonical figure 2026-09-01) credible: the same method, on terrain known not to have changed, says it didn't change. ### License CC-BY 4.0. Derivative work of PNOA-LiDAR (IGN/CNIG, Spain). Labeling context: © OpenStreetMap contributors (ODbL). See `METHOD.md` for the exact attribution. ## 简体中文 **困难负样本数据集 / 变化检测真值基准**：50个真实的、地形稳定的多边形区域—— 均为**激光雷达验证的"无变化"**样本，每个样本在两次真实的PNOA激光雷达（西班牙国家航空正射影像计划） 采集之间标注 `verdict=NO_CHANGE_DETECTED`（2019年11月25日 → 2024年4月16日， 日期通过激光雷达点云中嵌入的GpsTime核实，而非文件名中的名义日期）， 采用与测量Ca na Putxa垃圾填埋场（西班牙伊维萨岛）相同的物理方法生成。 ### 文件说明 - `polygons.geojson` — 包含全部属性的50个多边形，可直接导入QGIS/kepler.gl/Overpass turbo。 - `polygons.parquet` — 同一数据表的列式存储格式。 - `evidence/CTRL-*.json` — 每个已发布多边形对应一个证据文件（共50个，与数据表一一对应）。 - `passed_polygons.json` — 相同的50个样本，扁平JSON格式（同一数据表的替代格式）。 - `rejected_polygons.json` — 未通过覆盖率/阈值筛选的4,327个真实候选样本，完整数据及拒绝原因。 - `METHOD.md` — 详细方法说明，如实披露飞行日期方面的局限性，不做隐瞒。 **如实说明的修正（2026-09-01）**：原始脚本在抽取本次50个空间分散样本之前， 实际评估并通过筛选的候选样本共4,757个——但该4,757个样本的完整列表仅存在于运行时内存中， 运行结束后未保存到磁盘。未保存的数据无法发布：此处提供的是最终选定的50个样本 （完整数据及证据）和未通过筛选的4,327个样本（完整数据）——4,757这一数字作为该次运行的真实统计结果引用， 但并非可下载文件。 ### 为何重要 任何人都可以声称"我测量到了真实的体积变化"。但很少有人同时公开： 用同样的方法处理同样的数据，在何处会如实给出**"此处无变化信号"**的结论—— 采用相同的检测阈值、相同的最低覆盖率要求，未通过筛选的样本单独列出，而非隐藏。 这正是Ca na Putxa的测量结果（+310,357 ± 183立方米，2026-09-01标准数值）可信的原因： 同一方法用于已知未发生变化的地形，得出的结论也是"未变化"。 ### 许可协议 CC-BY 4.0。基于PNOA-LiDAR（西班牙国家地理研究院IGN/西班牙国家地理信息中心CNIG）数据的衍生作品。 标注上下文引用：© OpenStreetMap贡献者（ODbL协议）。详细署名方式见 `METHOD.md`。 ## Contact / 联系方式 NKO Data Labs — a Delaware LLC / 特拉华州有限责任公司。 contact@nkodatalabs.com https://nkodatalabs.com/zh-cn/products/verified-negatives/?utm_source=modelscope&utm_medium=dataset_card