A new benchmark, Infra-Bench CLS, has been introduced to evaluate the effectiveness of Earth observation foundation models in classifying critical infrastructure. The benchmark comprises 18,756 images of facility-scale infrastructure across seven continents, covering 13 classes. When compared to a ResNet-18 baseline, the best-performing foundation model achieved a 48% improvement in macro F1 scores, reaching 57.9%. While top classes like airports and data centers showed strong performance, sectors such as power infrastructure performed poorly, indicating areas for future research. AI
IMPACT This benchmark could accelerate the development and application of foundation models for critical infrastructure monitoring and analysis.
RANK_REASON The item describes a new academic benchmark and evaluation of foundation models for a specific task. [lever_c_demoted from research: ic=1 ai=1.0]
- AlphaEarth Foundations
- DINOv3 ViT-L/16
- Hugging Face
- Infra-Bench CLS
- OlmoEarth v1.1-Base
- Prithvi-EO-2.0
- ResNet-18
- SatlasPretrain S1
- SatlasPretrain S2
- Sentinel-1
- Sentinel-2
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →