OmDet

Why we included this project

OmDet-Turbo lets you detect objects without being locked into a fixed list of classes. The transformer-based detector takes free-form text as the query, so one model can find whatever categories you ask for at inference time, no retraining per label. It ships as an official PyTorch implementation with pretrained checkpoints on Hugging Face and ModelScope, plus a demo script and a small WSGI server if you want to wire it into existing services. The authors document ONNX export, which matters if you plan to run it on accelerated hardware, and the model is already merged into Hugging Face Transformers. For teams that need open-vocabulary detection to keep up with live video, it is a practical starting point.