LoveDA: 5,987 High-Resolution Remote Sensing Images for Land-Cover Segmentation

5,987 high spatial resolution (0.3 meter) remote sensing images from Nanjing, Changzhou, and Wuhan. The dataset was created by researchers from Wuhan University and RSIDEA for the 2021 NeurIPS Datasets and Benchmarks track. It focuses on land-cover classification across urban and rural environments to advance semantic segmentation and domain adaptation tasks.

Use Cases

• Training semantic segmentation models for land-cover classification based on high-resolution satellite imagery.
• Developing domain adaptation algorithms to handle shifts between urban and rural environments.
• Benchmarking model performance on challenges like multi-scale objects and complex backgrounds.
• Analyzing land-use patterns and changes in specific Chinese urban and rural areas.

Strengths

• 5,987 images provide a substantial collection for model training.
• High spatial resolution of 0.3 meters per pixel offers fine-grained detail.
• Explicitly designed for domain adaptation with data from three distinct cities.

Limitations

• Row count and total file size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

RSIDEA, Wuhan University.

Collection Method

Images sourced from Google Earth.

Geography

Nanjing, Changzhou, and Wuhan, China.