EarthDial: 10K-100K Evaluation Records for Remote Sensing VLMs

EarthDial-Dataset is a curated collection of 10,000 to 100,000 evaluation-only records for remote sensing and Earth observation, released by akshaydudhane and last updated in December 2024. It benchmarks vision-language models (VLMs) on real-world satellite and aerial imagery across tasks including classification, object detection, and change detection.

Use Cases

• Benchmarking vision-language models on remote sensing question answering
• Evaluating zero-shot object detection in aerial imagery
• Testing multimodal reasoning for change detection tasks

Strengths

• 10,000 to 100,000 record scale
• Multi-task coverage including change detection and QA
• Apache 2.0 licensed for open use

Limitations

• Evaluation-only dataset with no training or validation splits
• Geographic bias potentially limited to the US region per metadata tags
• Unknown specific sensor types or spatial resolutions

Provenance

Source

akshaydudhane via Hugging Face

Collection Method

Curated from remote sensing sources for evaluation purposes

Freshness

Last updated December 2024.

Geography

United States