T-Rex Label

T-Rex Label is an online annotation workspace designed to remove the slowest part of building a computer-vision dataset: drawing boxes by hand. Instead of labeling every instance, you give the model one visual or text prompt — for example a box drawn around a single object — and it finds and labels the matching objects across the image and, crucially, across other images in your batch. Because the underlying models are open-set, targets that were never part of a training set can still be detected without retraining.

The tool runs entirely in the browser, so there is nothing to install or deploy, which keeps onboarding light for teams and individual engineers alike. It supports common annotation operations including bounding boxes and segmentation, and imports and exports the formats most vision pipelines expect, so the resulting dataset can move into environments such as Kaggle, Roboflow, Hugging Face, Label Studio or FiftyOne without awkward conversion steps.

Supported models include open-vocabulary detectors in the Grounding DINO and T-Rex families. The platform is positioned for domains where labeling is expensive because targets are rare, visually varied or spread across large image sets — agriculture, livestock monitoring, electronics inspection, construction, retail and e-commerce, medical and life sciences, logistics and transportation.