Whenever you're ready, walk me through how you'd approach this.
Explain your deep learning domain knowledge for object detection.
In your answer, focus on how you would turn that knowledge into a production system: model choice, training data, inference pipeline, evaluation, and failure handling. Assume the system must support real-time detection with strict latency and accuracy requirements.