Aerial Photography Datasets, Covering Vehicle/ship Detection/object Assessment/cityscapes...
With the popularization of drones and the rapid development of computer vision technology, drone aerial photography, as an innovative photography method, is entering the public eye at an unprecedented speed. It breaks the limitations of traditional photography and opens up a "God's perspective" for us. The performance of aerial photography hardware is gradually approaching the physical limit, and the difficulty of algorithm optimization is also increasing.The quality of data directly determines whether the relevant model can move from simple data collection to accurate object assessment and scene classification.
The construction of an aerial photography dataset is by no means a simple stacking of images. Compared with traditional data collection methods, aerial photography can obtain large amounts of information data in a short period of time, greatly improving the efficiency of data collection. In addition, to ensure the validity and security of the data, it is necessary to rationally plan the collection area and time, strictly divide the training set, validation set, and test set, and establish a dynamic update mechanism to regularly supplement new data to adapt to the ever-changing geographical and physical environment. Faced with complex tasks such as urban planning, target detection, and object evaluation, when constructing a dataset, it is necessary to deeply analyze the needs of various fields, integrate multi-dimensional information, simulate real application scenarios, and provide practical learning materials for model training.
In short, the whole society is paying more and more attention to high-quality aerial photography datasets. Next, HyperAI has compiled a series of popular and practical aerial photography datasets from universities such as Tianjin University and Wuhan University.It involves multiple fields such as vehicle detection, ship detection, and object assessment.For practitioners and researchers who are committed to deepening their research in the field of aerial photography, these data sets are undoubtedly a great help.
Click to view more open source datasets:
Aerial photography dataset summary
1. FLAME fire aerial image dataset
Download address:https://go.hyper.ai/G1sUE
FLAME is a forest fire detection dataset based on aerial images, which aims to facilitate the monitoring and warning of forest fires. The data comes from fire images collected by drones during prescribed burning of piled debris in pine forests in Arizona, including video records and heat maps taken by infrared cameras.
2. SkyCity Aerial City Landscape Urban Landscape Aerial Photography Dataset
Estimated size:117.1 MB
Download address:https://go.hyper.ai/Zip37
This dataset is used for aerial landscape classification. It contains 8k images in total, including 10 different categories (bridges, commercial sites, industrial sites, intersections, landmarks, parks, parking lots, playgrounds, residential areas, and stadiums), and each category contains 800 high-quality images. The data sources include the public AID and NWPU-Resisc45 datasets, which are designed to facilitate urban landscape analysis.
Estimated size:154.31 MB
Download address:https://go.hyper.ai/loRNk
This dataset is used for aerial landscape classification. It contains 12k images in total, including 15 different categories (agriculture, airport, beach, city, desert, forest, grassland, road, lake, mountain, parking lot, port, railway, residential, river), each category contains 800 high-quality images with a resolution of 256×256 pixels. It aims to promote research and development in the field of computer vision, especially in aerial landscape analysis.
4. Ships/Vessels in Aerial Images
Estimated size:353.02 MB
Download address:https://go.hyper.ai/HGjn3
This dataset is dedicated to ship detection and contains a total of 26.9k images with bounding box annotations presented in YOLO format, which can achieve efficient and accurate ship detection, thus enabling a wide range of potential applications.
5. Bird vs Drone Bird and drone image classification dataset
Estimated size:1.05 GB
Download address:https://go.hyper.ai/fEhfo
The dataset comes from a variety of image collections on the Pexel website, containing a total of 20,925 images, captured as video frames, segmented, enhanced, and preprocessed to simulate different environmental conditions, and better identify drones and birds in various environments. The dataset is formatted according to the YOLOv7 PyTorch specification and is divided into 3 folders: Test, Train, and Valid.
*Test folder: Contains 889 drone and bird images. This folder has subcategories labeled BT (bird test images) and DT (drone test images).
*Train folder: This folder contains 18,323 images, including drone and bird images, divided into BT and DT categories.
*Valid folder: contains 1,740 images, and the folder images are divided into BT and DT.
6. iSAID Aerial Image Instance Segmentation Dataset
Estimated size:6.74 GB
Download address:https://go.hyper.ai/xZzWt
iSAID is the first benchmark dataset for instance segmentation in aerial images, combining instance-level object detection and pixel-level segmentation tasks. It contains 2,806 high-resolution images covering 15 categories and 655,451 object instances. The data comes from Google Earth, JL-1 satellite and GF-2 satellite (China Resources Satellite Data and Application Center).
7. DroneVehicle Large-scale drone aerial vehicle detection dataset
Estimated size:13.06 GB
Download address:https://go.hyper.ai/ZLJF0
This dataset was released by the Tianjin University research team in 2020. It contains 56,878 images, all of which are RGB images and infrared images. The images are from the research on vehicle detection and counting in drone aerial images. The five categories of car, turck, bus, van, and freight car are richly annotated with directional bounding boxes.
8. Spanish Traffic Aerial Image Dataset
Estimated size:32.3 GB
Download address:https://go.hyper.a/ERlyA
The data was captured by drones, with a total of 15,070 frames of images, covering a variety of traffic scenes such as regional roads, urban intersections, and rural roads. The creation process includes multiple steps such as data collection, image capture, vehicle annotation, anonymization, and data verification. It aims to provide high-quality training data for machine vision algorithms in the field of traffic management.
9. UAVid Aerial Photography Dataset
Estimated size:35.16 GB
Download address:https://go.hyper.ai/zESj9
This dataset is a high-resolution UAV semantic segmentation dataset, consisting of 30 video sequences, capturing 4K high-resolution images with oblique perspectives, and densely annotating 300 images with 8 classes for semantic labeling tasks. It has significant features in large-scale variation, moving target recognition, and temporal consistency preservation.
Estimated size:35.38 GB
Download address:https://go.hyper.ai/1JT9u
The DOTA-v1.0 dataset was published by Wuhan University on arXiv on November 28, 2017, and was later published at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) in June 2018. It contains 2,806 aerial images from Google Earth, JL-1 satellite, China Resources Satellite data, and GF-2 satellite. It can be used for target detection in aerial images, and for discovering and evaluating objects in images.
11. DOTA aerial image dataset v1.5+v2.0
Estimated size:53.12 GB
Download address:https://go.hyper.ai/0EIjk
The DOTA dataset is a collection of aerial images from different sensors and platforms, used for object detection in aerial images. This dataset is divided into v1.5 and v2.0 versions. The instances are marked by arbitrary quadrilaterals (8 degrees of freedom) by aerial image interpretation experts.
*DOTA-v1.5: Contains a total of 403,318 instances, adds annotations for extremely small instances (less than 10 pixels), and adds a new "container crane" category.
*DOTA-v2.0: There are 18 common categories, 11,268 images (divided into training, validation, test-dev and test-challenge sets) and 1,793,658 instances.
The above is the aerial photography dataset compiled by HyperAI. If you have resources that you want to include on the hyper.ai official website, you are welcome to leave a message or submit your contribution to tell us!