Embed the world: Multimodal AI for searchable aerial imagery at scale
Amazon Nova Multimodal Embeddings delivered highest F1 scores for natural-language semantic search over aerial imagery at scale.
AWS's GenAI Innovation Center partnered with aerial imagery provider Vexcel to build natural-language semantic search over multi-view aerial imagery using Amazon Nova Multimodal Embeddings, Bedrock, and OpenSearch Serverless. Their evaluation found Nova embeddings produced the best F1 scores, and the work became the Vexcel Intelligence product. It matters as a practical reference architecture for geospatial semantic search without per-feature model training.