Sat Image Boundingbox Sft Full: Satellite Imagery and Land-Cover Pairs for Fine-Tuning

Satellite imagery and aligned land-cover outputs packaged as image–text rows for fine-tuning in SFT format. The dataset, authored by PEBE691445, was last updated on April 28, 2026. JSONL user prompts name the modality where it matters.

Use Cases

• Fine-tuning vision-language models based on satellite imagery and text prompts.
• Training models for land-cover classification based on aligned optical and label data.
• Developing models for GeoGuessr-style location recognition based on stochastic POI data.
• Conducting research in multimodal remote sensing based on Sentinel-2 optical data.

Strengths

• Data is formatted for supervised fine-tuning (SFT) with image–text rows.
• Uses Sentinel-2 multispectral optical data from a public STAC catalog.
• Includes aligned land-cover outputs for training or validation.

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.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Sentinel-2 multispectral optical COGs from a public STAC catalog; POIs from stochastic/random_streetview_images_pano_v0.0.2.

Collection Method

Optical data percentile-stretched to uint8; locations described as GeoGuessr-style POIs.

Time Range

null

Freshness

Last updated 2026-04-28 05:38:44; freshness should be verified.

Geography

null