Edge AI or cloud: where should Computer Vision run?

There is no single correct architecture for Computer Vision. Processing video at the edge, on central servers or in the cloud involves trade-offs in latency, bandwidth, security, cost and availability.

When Edge AI makes sense

• Low latency matters.
• Connectivity is limited or unstable.
• Video volume makes cloud upload impractical.
• The operation must continue without external connectivity.

When central or cloud processing makes sense

• Sites have strong connectivity.
• Centralizing management and compute simplifies operations.
• The project needs elastic resource scaling.

Hybrid models

Many operations combine inference close to cameras with a central layer for events, dashboards, history and integrations.