computer-vision-engineering

Computer-vision engineering team — 3 agents for a production vision build: cv-systems-architect (task framing across classification / detection / segmentation / OCR / tracking / VLM, data & annotation strategy, build-vs-API and deployment-target choice, eval-metric design — mAP / IoU / precision-recall), cv-model-engineer (dataset curation, augmentation, transfer learning, model selection across YOLO / DETR / SAM / CLIP / ViT, active learning, the eval harness, drift), and vision-deployment-engineer (quantization / distillation, ONNX / TensorRT / CoreML / TFLite export, edge/embedded targets, streaming-video pipelines, latency). Vision-specific vs the MLOps-broad ml-engineering. Ships 4 skills, a knowledge bank (4 Mermaid decision trees + a dated 2026 reference), 5 best-practices, 2 templates, 2 commands. Engineering judgment, not a benchmark verdict; model/hardware/runtime specifics carry a retrieval date + [verify-at-use]; no PII, no image data stored. Requires ravenclaude-core@>=0.7.0.