Meta-Album RESISC: 800 Remote Sensing Images for 20 Scene Categories

NWPU-RESISC45 dataset provides 800 RGB images across 20 scene categories for remote sensing analysis. The images are 128x128 pixels and were preprocessed for the Meta-Album benchmark by Phan Anh VU in March 2022. The original dataset was created by Gong Cheng, Junwei Han, and Xiaoqiang Lu to provide a challenging benchmark with high within-class diversity and between-class similarity.

Use Cases

• Few-shot image classification based on 20 scene categories.
• Benchmarking model performance on remote sensing data with high intra-class variation.
• Training computer vision models for land cover and land use recognition from aerial imagery.

Strengths

• Contains 800 images, providing a defined scale for model training and evaluation.
• Images are uniformly preprocessed to a size of 128x128 pixels.
• The dataset is designed to be challenging with high within-class diversity and between-class similarity.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is a subset (Micro version) of the larger RESISC45 collection, which originally contained 45 categories and 31,500 images.
• License is CC-BY-NC 4.0, restricting commercial use.

Provenance

Source

NWPU-RESISC45 Dataset, preprocessed for Meta-Album.

Collection Method

Images were gathered and released by the original authors, then resized for the Meta-Album benchmark.