Meta-Album RESISC: 31,500 Satellite Images for 45 Scene Categories

Meta_Album_RESISC_Extended is a preprocessed version of the NWPU-RESISC45 dataset containing 31,500 RGB images across 45 remote sensing scene categories. The dataset was created by Gong Cheng, Junwei Han, and Xiaoqiang Lu and preprocessed for the Meta-Album benchmark by Phan Anh VU in March 2022. Images are sized at 128x128 pixels and are designed to be challenging due to high within-class diversity and between-class similarity.

Use Cases

• Training scene classification models based on 45 aerial scene categories.
• Benchmarking few-shot learning algorithms on a dataset with high within-class diversity.
• Developing computer vision models for land cover and land use analysis from satellite imagery.
• Studying model performance on a dataset with intentionally high between-class similarity.

Strengths

• Contains 31,500 images, providing a substantial volume for model training.
• Includes 45 distinct scene categories, offering a broad classification challenge.
• Images are preprocessed to a consistent 128x128 pixel size, simplifying input pipelines.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• The license (CC-BY-NC 4.0) prohibits commercial use, which may limit applicability.

Provenance

Source

NWPU-RESISC45 Dataset, preprocessed for the Meta-Album benchmark.

Collection Method

Likely gathered from satellite or aerial imagery sources, then resized and organized.