2,806 high-resolution aerial images containing 655,451 instances across 15 object categories such as ships, planes, and storage tanks. This dataset provides per-pixel instance segmentation masks and bounding box annotations for large-scale remote sensing imagery.
Use Cases
- Train instance segmentation models using the provided pixel-level masks and category labels
- Evaluate object detection algorithms using the bounding box coordinates for the 15 object classes
- Develop multi-scale feature extraction techniques to handle the significant variation in object sizes within the aerial images
Strengths
- 655,451 annotated instances across 15 distinct object categories including harbor, bridge, and helicopter
- Includes per-pixel instance segmentation masks and bounding box labels for every object
- Features high-resolution images with varying scales and orientations typical of remote sensing data