This pull request adds support for the Roboflow detection framework which includes provided object detection models from the Roboflow Universe (package inference) and local trained RFDETR models (package rfdetr).

This notebook is good as starting point: https://github.com/nok/sahi/blob/feature/rfdetr/demo/inference_for_roboflow.ipynb

Roboflow Model Integration:

• Added a new RoboflowDetectionModel class in sahi/models/roboflow.py, which includes methods for model initialization, dependency checks, inference, and converting predictions into the required format.
• Updated sahi/auto_model.py to register the new RoboflowDetectionModel under the key "roboflow" in the model registry.

Documentation Updates:

• Updated README.md to include installation instructions for the Roboflow detection framework, specifying the required packages inference>=0.50.3 and rfdetr>=1.1.0.

Testing

Unfortunately I couldn't add unit tests, because I got several dependency conflicts:

darius@ws:~/repos/sahi$ uv add "rfdetr>=1.1.0" "inference>=0.50.4" --group ci
  × No solution found when resolving dependencies for split (python_full_version >= '3.12' and sys_platform == 'darwin'):
  ╰─▶ Because matplotlib-stubs==0.2.0 depends on numpy>=1.20,<2.dev0 and only matplotlib-stubs<=0.2.0 is available, we can conclude that matplotlib-stubs>=0.2.0 depends on numpy>=1.20,<2.dev0.
      And because sahi:dev depends on matplotlib-stubs>=0.2.0 and inference==0.50.4 depends on numpy>=2.0.0,<2.3.0, we can conclude that inference==0.50.4 and sahi:dev are incompatible.

It would be great to get some help or advice if unit tests are required strictly.

Darius 👋