Model Cascades for Computer Vision: Building a Two-Stage Vision Pipeline

To reduce manual inspection without losing accuracy, build a two-stage model cascade in Roboflow Workflows. RF-DETR runs on every tile and accepts detections at and above 0.70 confidence, while the uncertain ones between 0.40 and 0.70 are cropped and sent to Gemini, and anything Gemini marks “unsure” goes to an inspector in Roboflow Annotate. Model cascades for computer vision connect models in stages, using an earlier model’s output to guide what a later model analyzes. In a conditional cascade, a fast model handles the initial pass, while selected images or regions receive additional processing. This gives teams a way to balance prediction quality, inference cost, and latency. With Roboflow Workflows, you can build a cascade that connects object detection, conditional routing, and a second model in one pipeline. For example, a manufacturer inspecting ceramic tiles can use RF-DETR (opens in a new tab) to identify potential defects, send uncertain regions to Gemini (opens in a new tab) for verification, and route unresolved cases to an inspector. RF-DETR is released under Apache 2.0, so it's free for commercial use. This guide explains how vision model cascades work, when to use them, and how to choose routing rules and second-stage inputs. You’ll…