datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
WHU-Building-Dataset
WHU Building Dataset
The WHU Building Dataset is a widely-used benchmark for building extraction from high-resolution aerial imagery. It contains aerial images at 0.3m resolution with pixel-level binary building masks.
Dataset Description
Property
Value
Resolution
0.3m ground sampling distance
Tile Size
512 x 512 pixels
Channels
3 (RGB)
Classes
2 (Background=0, Building=255)
Format
PNG
Splits
Split
Images
Masks… See the full description on the dataset page: https://huggingface.co/datasets/giswqs/WHU-Building-Dataset.drone-building-scans
Drone Building Scans
Two commercial buildings in the Minneapolis suburbs, flown August 2026. Full image sets with
GPS intact, solved camera poses, and the finished 3D models.
Free drone photogrammetry test data for buildings barely exists — published datasets are
either a whole city block or a turntable object, with nothing in between.
CC BY 4.0.
⬇ Download
→ huggingface.co/datasets/Matt1up/drone-building-scans
Browse the Files tab and take what you… See the full description on the dataset page: https://huggingface.co/datasets/Matt1up/drone-building-scans.satellite-building-segmentation
Dataset Labels
['building']
Number of Images
{'train': 6764, 'valid': 1934, 'test': 967}
How to Use
Install datasets:
pip install datasets
Load the dataset:
from datasets import load_dataset
ds = load_dataset("keremberke/satellite-building-segmentation", name="full")
example = ds['train'][0]
Roboflow Dataset Page
https://universe.roboflow.com/roboflow-universe-projects/buildings-instance-segmentation/dataset/1
Citation… See the full description on the dataset page: https://huggingface.co/datasets/keremberke/satellite-building-segmentation.vhr-building-segmentation
HOT Building Segmentation Dataset
Dataset Description
A semantic segmentation dataset for building footprint extraction from aerial imagery, built from validated Humanitarian OpenStreetMap Team (HOT) Tasking Manager projects that use OpenAerialMap (OAM) imagery.
Dataset Summary
This dataset pairs 256x256 aerial image tiles (zoom level 19) from OpenAerialMap with building footprint labels from OpenStreetMap. All source projects have been fully… See the full description on the dataset page: https://huggingface.co/datasets/hotosm/vhr-building-segmentation.arch-building-dataset
World Architectural Buildings Dataset (FGIC) for Multi‑Class Image Classification
Multi‑Class Image Classification dataset of world architectural buildings with finalized curation.
Classes
Class
Count
Description
barn
1,680
Traditional wooden barn architecture — residential and storage buildings
bridge
1,680
Various bridge architectures (suspension, arch, truss)
castle
1,680
Medieval and modern castle structures
mosque
1,680
Islamic mosque… See the full description on the dataset page: https://huggingface.co/datasets/0xgr3y/arch-building-dataset.DisasterM3_road_building_Opticalhot-building-segmentation
HOT Building Segmentation Dataset
Dataset Description
A semantic segmentation dataset for building footprint extraction from aerial imagery, built from validated Humanitarian OpenStreetMap Team (HOT) Tasking Manager projects that use OpenAerialMap (OAM) imagery.
Dataset Summary
This dataset pairs 256x256 aerial image tiles (zoom level 19) from OpenAerialMap with building footprint labels from OpenStreetMap. All source projects have been fully validated through… See the full description on the dataset page: https://huggingface.co/datasets/kshitijrajsharma/hot-building-segmentation.marburg-university-buildings
Marburg University Buildings
The Marburg University Buildings dataset contains high-resolution photographs of 29 university buildings in Marburg, Germany. Each image is labeled by the name of the building and intended for use in multi-class image classification tasks.
Dataset Summary
Total classes: 29
Image format: JPG
Label format: Folder name as string
Collection: Manually photographed under consistent daylight and angle conditions (SAMSUNG galaxy A54)
No compression… See the full description on the dataset page: https://huggingface.co/datasets/soroushdft/marburg-university-buildings.drone-building-extraction
🏘️ Drone Building Extraction Dataset — Calgary
A high-resolution drone imagery dataset for binary building segmentation,
captured over a residential area in Calgary, Alberta, Canada.
Designed to work directly with GeoSeg Studio
— an open-source QGIS plugin for deep learning semantic segmentation.
🖼️ Preview
Train Area (red = building polygons)
Test Area (blue = building polygons)
📋 Dataset Summary
Property
Value
Location
Calgary… See the full description on the dataset page: https://huggingface.co/datasets/dronnix-io/drone-building-extraction.BuildingTypeBuilding-Urban-Climate
Urban Climate Dataset
Dataset Description
This dataset is part of the Urban Climate project. It contains satellite images of urban areas and corresponding segmentation masks that are used for building detection and segmentation tasks.
Files and Structure:
train/: Contains training images and masks.
train/Image/: Original satellite images used for training.
train/Mask/: Binary masks where buildings are labeled.
validation/: Contains validation images and… See the full description on the dataset page: https://huggingface.co/datasets/nave1616/Building-Urban-Climate.nfu_buildingconfigs:
config_name: default
data_files:
split: train
path: data/train-*
diffusion_skyscrapers_city_building
A Dataset of skyscrapers in the style of a city building game
Terms and Conditions
The dataset is usable for training and experimenting with new models, but since the images are generated using OpenAI's API, as stated by them, if your project involves sensitive or controversial content, you should review their policies and terms to ensure compliance
https://openai.com/it-IT/policies/usage-policies
https://openai.com/it-IT/policies/terms-of-use
If you copy and distribute… See the full description on the dataset page: https://huggingface.co/datasets/Loulblemo/diffusion_skyscrapers_city_building.building_height_estimation
Building Contour Detection and Height Estimation Problem
Dataset Summary
The building_height_estimation dataset is a collection of satellite images with annotated building footprints (polygons) and corresponding building heights.
It is designed for the joint tasks of building contour detection (segmentation) and height estimation (regression) from monocular aerial images.
Source & Owner: The dataset originates from the AlgoTester Building Contour Detection & Height… See the full description on the dataset page: https://huggingface.co/datasets/MElHuseyni/building_height_estimation.Greater-Sydney-Trees-BuildingsExplore our Greater Sydney Building Footprint and Tree Patch datasets using a PMTiles map here
aigis
AIGIS
AI annotation, segmentation, and conversion tools for GIS imagery
aigis is a comprehensive toolkit for aerial and satellite imagery acquisition, processing, annotation, and analysis using artificial intelligence. This repository contains three main components:
annotate: Tools for annotating aerial imagery data.
convert: Utilities for converting… See the full description on the dataset page: https://huggingface.co/datasets/SIH/Greater-Sydney-Trees-Buildings.satellite-building-segmentation
Dataset Labels
['building']
Number of Images
{'train': 6764, 'valid': 1934, 'test': 967}
How to Use
Install datasets:
pip install datasets
Load the dataset:
from datasets import load_dataset
ds = load_dataset("keremberke/satellite-building-segmentation", name="full")
example = ds['train'][0]
Roboflow Dataset Page
https://universe.roboflow.com/roboflow-universe-projects/buildings-instance-segmentation/dataset/1
Citation… See the full description on the dataset page: https://huggingface.co/datasets/merve/satellite-building-segmentation.spacenet-buildings-vegas-smoketest-sample
SpaceNet Buildings (Vegas) — 20-Chip Smoke-Test Sample
This is a tiny 20-chip sample, not a full dataset. It exists to reproducibly
smoke-test a U-Net vs YOLO building-detection training pipeline end-to-end — not to
train a real model. See the companion model repo
khalilurrahmanridoykhan/unet-vs-yolo-buildings-smoketest
and the source code at
github.com/khalilurrrahmanridoykhan/best-model-for-satellite-imagery.
Source and license
Derived from SpaceNet Buildings v2… See the full description on the dataset page: https://huggingface.co/datasets/khalilurrahmanridoykhan/spacenet-buildings-vegas-smoketest-sample.vhr-buildings-tidied-v2-lower-rankcomplete-buildings-extraction-coco-hfbuildingstyles
Building Styles Dataset
This dataset contains images of buildings labeled with different architectural styles. Each image has a corresponding text prompt.
building-15001.5k image filtered by most damage coverage from xBD dataset
building_segmentationbuilding_512pc-building-egocentricbuildingsbuilding_defect_vqa
Building Defect VQA Dataset
Dataset Description
This dataset is a Visual Question Answering (VQA) version of the original BD3 (Building Defect Dataset).
It is designed for training and evaluating Vision–Language Models (VLMs) on building defect recognition tasks.
Each image is paired with a fixed question and a defect category as the answer.
Dataset Structure
Each sample contains:
image: RGB image of a building surface
question:
"What type of building… See the full description on the dataset page: https://huggingface.co/datasets/chandrabhuma/building_defect_vqa.PhilEOBench-building_density_regression
Simulated PhiSat Bench Dataset - Buildings
This repository contains a simulated dataset derived from Sentinel-2 data for building analysis.
Specifically, the dataset simulates outputs from the PhiSat2 satellite.
Label Description
Each sample in the dataset includes a single-channel label.
The labels are stored as floating-point values that represent the estimated percentage of building coverage within each pixel.
For a pixel with a 10-meter resolution… See the full description on the dataset page: https://huggingface.co/datasets/ESA-PhiLab-Edge/PhilEOBench-building_density_regression.vhr-buildings-tidied-v2morocco-public-buildings-water-consumption
🇲🇦 Morocco Public Buildings — Water Consumption Dataset
Synthetic dataset for predicting annual water consumption (Conso_m3) in Moroccan public buildings across all 12 administrative regions.
📊 Dataset Overview
Property
Value
Rows
5,000 buildings
Target
Conso_m3 (m³/year)
Features
7 (2 categorical, 3 continuous, 2 binary)
Regions
12 Moroccan administrative regions
Climate zones
4 (Méditerranéen, Semi-aride, Semi-aride chaud, Désertique)
Building… See the full description on the dataset page: https://huggingface.co/datasets/Mauri789/morocco-public-buildings-water-consumption.buildings-extraction-coco-hf
Building Extraction Dataset
This dataset is a processed verison of the dataset of the Kaggle competition: https://www.kaggle.com/competitions/building-extraction-generalization-2024/.
The original train and validation images and COCO annotations were resized to (512, 512).
Then from the segmentations on the COCO annotations, a PIL_annotation file was created for each sample in which the Red channel corresponds to the semantic segmentation mask and the Green channel to the instance… See the full description on the dataset page: https://huggingface.co/datasets/tomascanivari/buildings-extraction-coco-hf.