Train/val split manifests extracted from GeoChat_Instruct for fine-tuning vision-language models on satellite scene classification. This Hub dataset includes a real image column so the dataset viewer shows the image, question, and answer aligned row-by-row.

Classes

45 scene categories (e.g. airplane, baseball diamond, beach, bridge, …)

	Splits

Split Samples

train ~25 200

val ~4 725

Format

Columns: image (previewable… See the full description on the dataset page: https://huggingface.co/datasets/tarek199147/nwpu-resisc45-v1.