Benchmark RadarAI BENCHMARK PROFILE
OVEarth-Bench
Evaluates open-vocabulary Earth observation models on category breadth and query diversity, covering mask and box localization across vocabulary, referring, and reasoning queries under a unified zero-shot protocol.
- Released
- 2026-07-29
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Existing EO benchmarks cover limited categories and query forms, making it hard to gauge real-world capability. This benchmark provides a broader, more diverse evaluation to compare general and EO-specific models on open-vocabulary localization.
Motivation
Open-vocabulary Earth observation (EO) aims to localize geospatial concepts specified in natural language rather than a fixed label set.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.