five

IRSC/ELDOR-sample

收藏
Hugging Face2026-04-19 更新2026-04-26 收录
下载链接:
https://hf-mirror.com/datasets/IRSC/ELDOR-sample
下载链接
链接失效反馈
官方服务:
资源简介:
--- configs: - config_name: site data_files: - split: sample path: site/metadata.csv default: true - config_name: patches data_files: - split: sample path: patches/metadata.csv pretty_name: ELDOR Sample task_categories: - image-segmentation tags: - remote-sensing - semantic-segmentation - uav - orthomosaic --- # ELDOR Sample This repository is a compact sample companion to the full `ELDOR` dataset: - Full dataset: `https://huggingface.co/datasets/IRSC/ELDOR` - It is intended for quick inspection of the imagery, labels, and released patch extraction without downloading the full release. ## What this sample contains This sample contains one representative site and all of its released patches: - Representative site: `AcumulacionAaron2B` (`train` split) - Site-level aligned orthomosaic image and pixel-aligned label map - All released `512x512` patches derived from that site with `256` stride and the public crop filter used in ELDOR Two dataset configurations are exposed: - `site`: the full aligned site-level image/label pair - `patches`: every released patch derived from that site ## How the sample was created The sample was created deterministically from the public ELDOR release: 1. We selected `AcumulacionAaron2B` because it has the broadest per-site class coverage in the public release: 12 of the 14 foreground classes. 2. We copied the exact released aligned orthomosaic PNG and label PNG for that site from the full dataset. 3. We copied every released patch derived from that site from the public `data-cropped` export without additional filtering beyond the original ELDOR crop pipeline. The public crop pipeline used for ELDOR is: - patch size: `512x512` - stride: `256` - full windows only, no padding - drop a patch if background exceeds `80%` - patch images stored as JPEG and patch labels stored as PNG This sample includes the following foreground classes: - Building - Mining raft - Primary Forest - Heavy machinery - Water bodies - Agricultural crop - Gravel mounds - Type 1 natural regeneration - Type 2 natural regeneration - Bare ground - Sluice - Vehicles Foreground classes present in the full dataset but not in this single-site sample: - Compact mounds - Grass ## Sample size summary - Site subset: 1 aligned image/label pair, `1.481` GB - Patch subset: `9849` image patches and `9849` label patches, `1.352` GB - Total sample size: `2.832` GB ## File organization ### `site` configuration - `site/images/AcumulacionAaron2B.png` - `site/labels/AcumulacionAaron2B.png` - `site/metadata.csv` ### `patches` configuration - `patches/images/AcumulacionAaron2B_<row>_<col>.jpg` - `patches/labels/AcumulacionAaron2B_<row>_<col>.png` - `patches/metadata.csv` Patch filename indices are 1-based row and column indices in the released crop grid.

配置项: - 配置名称:site(站点) 数据文件: - 拆分集:sample(样本集) 路径:site/metadata.csv 设为默认配置 - 配置名称:patches(分块) 数据文件: - 拆分集:sample(样本集) 路径:patches/metadata.csv 数据集展示名称:ELDOR样本集 任务类别: - 图像分割(image-segmentation) 标签: - 遥感(remote-sensing) - 语义分割(semantic-segmentation) - 无人机(UAV) - 正射影像(orthomosaic) # ELDOR样本集 本仓库为完整`ELDOR`数据集的轻量化配套样本集: - 完整数据集地址:`https://huggingface.co/datasets/IRSC/ELDOR` - 本样本集用于快速查看影像、标注以及提取分块,无需下载完整数据集。 ## 本样本集包含的内容 本样本集包含一个典型研究站点及其所有公开发布的分块: - 典型研究站点:`AcumulacionAaron2B`(对应`train`拆分集) - 站点级对齐正射影像(orthomosaic)与像素对齐标注图 - 所有基于该站点生成的公开`512×512`分块,采用`256`步长,且使用了`ELDOR`数据集的公开裁剪过滤规则。 ## 数据集配置模式 本数据集提供两种配置模式: - `site`:完整的站点级对齐影像-标注对 - `patches`:从该站点导出的所有公开分块 ## 样本集的生成方式 本样本集严格基于公开版`ELDOR`数据集确定性生成: 1. 我们选择`AcumulacionAaron2B`,因其在公开版数据中拥有最全面的单站点类别覆盖:包含14个前景类别中的12个。 2. 我们从完整数据集中复制该站点的精确对齐正射影像PNG与标注PNG文件。 3. 我们从公开`data-cropped`导出文件中复制所有该站点对应的公开分块,仅保留`ELDOR`原始裁剪流程的过滤规则,未添加额外筛选条件。 `ELDOR`数据集采用的公开裁剪流程如下: - 分块尺寸:`512×512` - 步长:`256` - 仅保留完整窗口,无填充 - 若分块中背景占比超过`80%`则丢弃该分块 - 分块影像存储为JPEG格式,分块标注存储为PNG格式 本样本集包含以下前景类别: - 建筑(Building) - 采矿筏(Mining raft) - 原始森林(Primary Forest) - 重型机械(Heavy machinery) - 水体(Water bodies) - 农作物(Agricultural crop) - 砾石堆(Gravel mounds) - 1型自然更新区(Type 1 natural regeneration) - 2型自然更新区(Type 2 natural regeneration) - 裸地(Bare ground) - 洗矿槽(Sluice) - 车辆(Vehicles) 完整数据集中存在但本单站点样本集未包含的前景类别: - 紧凑型土堆(Compact mounds) - 草地(Grass) ## 样本集规模汇总 - 站点子集:1组对齐影像-标注对,占用`1.481` GB - 分块子集:`9849`张影像分块与`9849`张标注分块,占用`1.352` GB - 总样本大小:`2.832` GB ## 文件组织结构 ### `site` 配置模式 - `site/images/AcumulacionAaron2B.png` - `site/labels/AcumulacionAaron2B.png` - `site/metadata.csv` ### `patches` 配置模式 - `patches/images/AcumulacionAaron2B_<row>_<col>.jpg` - `patches/labels/AcumulacionAaron2B_<row>_<col>.png` - `patches/metadata.csv` 分块文件名的行、列索引为公开裁剪网格中的1起始索引。
提供机构:
IRSC
5,000+
优质数据集
54 个
任务类型
进入经典数据集
二维码
社区交流群

面向社区/商业的数据集话题

二维码
科研交流群

面向高校/科研机构的开源数据集话题

数据驱动未来

携手共赢发展

商业合作