restor/tcd

OAM-TCD是一个全球多样化的高分辨率树冠覆盖地图数据集，包含28万棵树木和5.6万棵树群的实例级掩码。数据集中的图像以2048x2048像素的RGB GeoTIFF瓦片形式提供，可用于训练实例分割模型和语义分割模型。数据集由Restor / ETH Zurich策划，并由Google.org AI for Social Good资助。数据集的主要用途是映射无人机或航空调查中的树冠覆盖。数据集的结构包括图像、语义掩码和对象分割（实例多边形），并提供了OAM的元数据。数据集的创建过程包括从OpenAerialMap下载图像、注释过程以及注释者的补偿策略。数据集还包含地理分布和生物群落分布的统计信息，并讨论了数据集的偏见、风险和限制。

The OAM-TCD dataset is a globally diverse dataset of high-resolution tree cover maps, containing instance-level masks for 280k trees and 56k tree groups. The images are 2048x2048 px RGB GeoTIFF tiles sourced from OpenAerialMap (OAM), part of the Open Imagery Network (OIN). The dataset supports both instance segmentation and semantic segmentation tasks. It is curated by Restor / ETH Zurich and funded by Restor / ETH Zurich and a Google.org AI for Social Good grant. The dataset is licensed under CC-BY 4.0, with some images under less permissive licenses like CC BY-NC 4.0 or CC BY-SA 4.0, which are distributed in separate repositories. The dataset includes various features such as image_id, image, height, width, annotation, oam_id, license, biome, crs, bounds, validation_fold, biome_name, lat, lon, segments, meta, and coco_annotations. The dataset is split into train and test sets with 4169 and 439 examples respectively. The dataset is intended for mapping tree cover in aerial orthomosaics and can be used for training models, benchmarking, and direct applications where RGB input images are processed to produce tree (canopy) maps. However, it is not suitable for applications requiring detailed annotations of trees in closed canopy or for carbon sequestration measurement without additional structural or species information.

搜集汇总

背景与挑战

背景概述

OAM-TCD是一个高分辨率（10厘米/像素）的树木覆盖地图数据集，包含28万棵树木和5.6万棵树群的实例级掩码，图像为2048x2048像素的RGB GeoTIFF图块，适用于训练实例分割和语义分割模型。数据集来源自OpenAerialMap，具有全球多样性，覆盖多种陆地生物群落，但存在地理和注释偏差，主要用于树木覆盖检测和生态研究应用。