Crop Regions of Interest & Stitch Them Before VLM Inference

To cut VLM requests without losing detail, use an object detection model to find each region of interest, crop those regions, and stitch them into one numbered collage, so a VLM like Gemini can review up to ten at once and return a finding for each. Roboflow Workflows lets you build this full pipeline (RF-DETR detection, cropping, collage stitching, and structured Gemini reports) in one place, so you can start free and measure the token savings on your own images. Manufacturing quality inspection often comes down to small details. A torn carton edge, a punctured panel, or a crushed corner can reveal packaging damage that needs closer review. An inspection camera captures the whole package, while the evidence needed to assess its condition may occupy only a small part of the image. An object detection model (opens in a new tab) can locate these areas, and a vision-language model (VLM) (opens in a new tab) can describe the visible evidence in each one. As more packages move through inspection, sending every detected region in a separate VLM request repeats the prompt and adds more images to process. Cropping the regions of interest (ROIs) and stitching them into a numbered collage…