NWPU VHR-10 dataset

Dataset Overview

Dataset Name: VHR-10_dataset_coco

Dataset Description: VHR-10_dataset_coco is an object detection and instance segmentation dataset for very‑high‑resolution (VHR) remote‑sensing images. It is built on the NWPU VHR‑10 dataset and annotated following the COCO format. The original NWPU VHR‑10 dataset includes 10 object categories across 800 images; 715 color images are from Google Earth (0.5–2 m resolution) and 85 color‑infrared images are from Vaihingen (0.08 m resolution).

Dataset Composition:

• Positive set: 650 images containing at least one target.
• Negative set: 150 images containing no targets.

Annotation Information:

• In the positive set, 757 aircraft, 302 ships, 655 storage tanks, 390 baseball fields, 524 tennis courts, 159 basketball courts, 163 athletics fields, 224 ports, 124 bridges and 477 vehicles are manually annotated with bounding boxes and instance masks.

Dataset Structure:

├── show_coco.py # visualization script
├── NWPU VHR-10_dataset_coco
    ├── 正样本集 # 650 images
       ├── [.jpg]
       ├── ...
    ├── annotations.json # 650 annotations
    ├── split_datasets.py # random split script

Dataset Download:

• Download link: NWPU VHR-10 Dataset

Citation Information

If you use this dataset, please cite it as follows:

[1] Hao Su, Shunjun Wei, Min Yan, Chen Wang, Jun Shi, Xiaoling Zhang, "OBJECT DETECTION AND INSTANCE SEGMENTATION IN REMOTE SENSING IMAGERY BASED ON PRECISE MASK R-CNN," IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium, Yokohama, 2019.