awesome-satellite-imagery-datasets presents a curated list of datasets containing satellite and aerial imagery with associated annotations. The primary goal is to provide a centralized resource for computer vision and deep learning researchers working with remote sensing data. This repository addresses the challenge of finding suitable, well-annotated datasets for training and evaluating models in various applications such as object detection, segmentation, and scene classification.
This repository offers a comprehensive collection encompassing various data types and annotation formats. It features datasets from both academic challenges and commercial initiatives, with a focus on recent and high-quality resources. The list is regularly updated, and a clear structure allows for easy browsing by task type (e.g., Instance Segmentation, Object Detection).
- Instance Segmentation: Datasets with pixel-level annotations for identifying and segmenting objects within satellite imagery (e.g., PASTIS, SpaceNet 7).
- Object Detection: Datasets focused on bounding box annotations for detecting specific objects (e.g., iSAID, xView 2, Airbus Building Footprints).
- Building Footprints: Datasets providing building outlines with varying resolutions and sizes (e.g., Microsoft BuildingFootprints, SpaceNet 4).
- Scene Classification: Datasets for classifying entire satellite images into predefined categories (e.g., Agriculture-Vision Database).
- Human Activity: Datasets for annotating human activities from satellite imagery.
- Disaster Assessment: Datasets used for evaluating models that can assess damage and features after disaster(e.g., xView2 Building Damage Asessment Challenge).
- Aerial/Satellite Image Challenges: Links to competition websites offering complete datasets and evaluation metrics.
The repository is actively maintained, although it is currently archived. It was initially updated frequently, and a wide selection of datasets are provided. Recent additions highlight ongoing efforts in the field. Given the variety of sources and data types, the maturity varies between individual datasets, with some being well-established and others being more recent.
This resource benefits researchers, developers, and students interested in applying computer vision and deep learning to satellite imagery. It is valuable for projects involving land cover analysis, urban planning, disaster response, environmental monitoring, and various other applications. By providing curated datasets, it reduces the effort required to find and prepare data, accelerating research and development.