RSHR-Bench: A Benchmark for Multimodal LLMs on Ultra-High-Resolution Remote Sensing Data

RSHR-Bench is a benchmark dataset for evaluating Multimodal Large Language Models (MLLMs) using ultra-high-resolution remote sensing data. The dataset was created by RL-MIND and a Hugging Face Parquet version with embedded images was released on May 19, 2026. The accompanying paper 'RSHR: A Benchmark for MLLMs on Ultra-High-Resolution Remote Sensing Data' was released in November 2025.

Use Cases

• Benchmarking MLLM performance on ultra-high-resolution remote sensing imagery.
• Training models for visual question answering tasks based on satellite or aerial imagery.
• Developing models for geospatial scene understanding and description based on high-resolution data.

Strengths

• Focuses on ultra-high-resolution remote sensing data, a specific and challenging data type.
• Released in a Hugging Face Parquet format with embedded images, which may simplify data loading.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.

Provenance

Source

RL-MIND

Freshness

Last updated 2026-05-19 10:22:48.