Meta-Album RSD: Remote Sensing Scene Images for Few-Shot Learning

1520 images across 38 scene categories, resized to 128x128 pixels, from the RSD46 remote sensing collection. The dataset was preprocessed for the Meta-Album benchmark by Phan Anh VU in March 2022. Original images were collected from Google Earth and Tianditu by researchers at Wuhan University.

Use Cases

• Benchmarking few-shot image classification models based on remote sensing scene categories.
• Training models for land cover and land use classification based on satellite and aerial imagery.
• Developing domain adaptation techniques for the remote sensing domain using a standardized dataset format.

Strengths

• Contains 1520 images, providing a standardized subset for few-shot learning experiments.
• Images are uniformly preprocessed to a 128x128 pixel resolution.
• Includes 38 distinct scene categories for multi-class classification tasks.

Limitations

• The dataset is a mini version; the original RSD46 dataset contains 117,000 images, suggesting this is a significantly smaller sample.
• Description metadata is limited; actual data quality and class balance require manual inspection after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

RSD46-WHU (Wuhan University), preprocessed for Meta-Album.

Collection Method

Images collected from Google Earth and Tianditu, then resized and formatted for the benchmark.