SegFly
收藏资源简介:
SegFly是一个大规模航拍语义分割数据集,旨在支持RGB-热成像多模态语义分割研究。数据集包含20,606张高分辨率RGB图像和15,007组像素对齐的RGB-热成像图像对,总计35,613个样本。数据采集于多样化的环境,并在三种不同的无人机飞行高度(30米、40米、50米)下进行。数据集提供了精细的语义标注,涵盖15个基准类别,包括道路、人行道、泥土、砾石、草地、植被、树木、地面障碍物、车辆、水体、建筑物、屋顶、停车场、施工区域和卡车。数据按照场景和传感器模态(RGB或热成像)进行了划分,包含训练集、验证集和测试集。每个数据样本包含原始图像、语义分割标签、场景标识符、飞行高度和模态信息。对于热成像样本,还提供了配准后的RGB图像。该数据集适用于计算机视觉领域的语义分割任务,特别是多模态(RGB与热成像)融合、航拍图像理解、遥感分析以及相关算法的开发与评估。
SegFly is a large-scale aerial semantic segmentation dataset designed to support RGB-thermal (RGB-T) multimodal semantic segmentation research. The dataset includes 20,606 high-resolution RGB images and 15,007 pixel-aligned RGB-thermal image pairs, totaling 35,613 samples. Data was collected in diverse environments and at three different drone flight altitudes (30 meters, 40 meters, 50 meters). The dataset provides fine-grained semantic annotations covering 15 benchmark categories, including road, sidewalk, dirt, gravel, grass, vegetation, tree, ground obstacle, vehicle, water body, building, roof, parking lot, construction area, and truck. Data is divided based on scene and sensor modality (RGB or thermal), containing training, validation, and test sets. Each data sample includes the original image, semantic segmentation labels, scene identifier, flight altitude, and modality information. For thermal samples, registered RGB images are also provided. The dataset is suitable for semantic segmentation tasks in computer vision, particularly for multimodal (RGB and thermal) fusion, aerial image understanding, remote sensing analysis, and the development and evaluation of related algorithms.
SegFly 数据集概述
SegFly 是一个大规模航空语义分割数据集,专注于 RGB 和热红外(RGB-T)多模态场景理解,总样本数为 35,613(20,606 张 RGB 图像 + 15,007 张热红外图像),图像分辨率为高分辨率,采集自 9 个不同场景 和 三种飞行高度(30m、40m、50m)。
数据集结构
| 特征 | 类型 | 描述 |
|---|---|---|
image |
Image |
原始传感器帧(RGB 或长波红外热成像) |
label |
Image |
8 位单通道语义掩码,映射到 15 个基准类别 |
RGB_aligned |
Image |
已配准的 RGB 图像(仅热红外模态提供;RGB 模态返回 None) |
scene |
string |
场景标识符("scene_01" 到 "scene_09") |
altitude |
string |
飞行高度("30m"、"40m"、"50m") |
modality |
string |
传感器模态("RGB" 或 "thermal") |
数据集划分与统计
| 模态 | 划分 | 场景 | 样本数 |
|---|---|---|---|
| RGB | 训练集 | scene_01 ~ scene_05 |
14,738 |
| 验证集 | scene_06、scene_07 |
1,965 | |
| 测试集 | scene_08、scene_09 |
3,842 | |
| 热红外 | 训练集 | scene_03 ~ scene_05 |
12,063 |
| 验证集/测试集 | scene_09 |
2,944 |
类别映射(15 个基准类 + 未标注类)
| 类别 ID | 类别名称 | RGB 颜色 |
|---|---|---|
| 0 | 未标注/忽略 | [0, 0, 0] |
| 1 | 道路 | [128, 0, 128] |
| 2 | 人行道 | [204, 163, 72] |
| 3 | 泥土 | [128, 0, 0] |
| 4 | 砾石 | [192, 192, 192] |
| 6 | 草地 | [0, 255, 0] |
| 7 | 植被 | [112, 148, 32] |
| 8 | 树木 | [64, 64, 0] |
| 9 | 地面障碍物 | [255, 255, 0] |
| 13 | 车辆 | [0, 128, 128] |
| 14 | 水体 | [0, 0, 255] |
| 16 | 建筑物 | [255, 0, 0] |
| 17 | 屋顶 | [64, 160, 120] |
| 33 | 停车场 | [128, 64, 128] |
| 34 | 施工区域 | [240, 120, 120] |
| 36 | 卡车 | [128, 128, 64] |
使用方式
通过 Hugging Face datasets 库加载:
python from datasets import load_dataset
dataset = load_dataset("markus-42/SegFly")
引用信息
bibtex @inproceedings{gross2026segfly, title={{SegFly: A Dataset and 2D-3D-2D Paradigm for Aerial RGB-Thermal Semantic Segmentation at Scale}}, author={Markus Gross and Sai Bharadhwaj Matha and Rui Song and Viswanathan Muthuveerappan and Conrad Christoph and Julius Huber and Daniel Cremers}, booktitle = {Proceedings of the European Conference on Computer Vision (ECCV)}, year={2026}, }
许可协议
采用 CC BY-NC-SA 4.0 许可协议。




