Design an image detection system

Design an end-to-end image object-detection system that ingests user images, runs detection models, and serves results via APIs. Specify functional/non-functional requirements (accuracy, latency, throughput, availability), high-level architecture (ingestion, storage, preprocessing, model serving, asynchronous workers/queues), data/version management, and how you would handle batching, GPU utilization, autoscaling, and caching. Describe model choices (single-stage vs. two-stage), training/labeling pipeline, evaluation metrics, online/offline monitoring, A/B testing strategy, failure modes/backpressure, privacy/compliance, cost controls, and rollback/blue-green deployments.