Delineate Anything YOLO Model

[isaaccorley]

This PR adds the Delineate Anything YOLO11x and YOLO11n pretrained models. This adds a models extras group to the dependencies to add the ultralytics package. The models are available as AGPL-3.0 licensed.

Note: the model is trained on a variety of satellite imagery with different resolutions so it's "spatial agnostic"

Sample predictions on FTW imagery:

[adamjstewart]

@robmarkcole wanted to check with you and see how you feel about this. Ultralytics would not be a required dependency, so this change shouldn't impact you.

[robmarkcole]

@adamjstewart I think using optional imports is a good way to go

[adamjstewart]

@isaaccorley might need to find a way to update this error message:

ultralytics is not installed and is required to use this dataset. Either run:

$ pip install ultralytics

to install just this dependency, or:

$ pip install torchgeo[datasets]

to install all optional dataset dependencies.

[isaaccorley]

@isaaccorley might need to find a way to update this error message:

ultralytics is not installed and is required to use this dataset. Either run:



$ pip install ultralytics



to install just this dependency, or:



$ pip install torchgeo[datasets]



to install all optional dataset dependencies.

Yeah I saw this. I might save this for another PR because I'm going to add a group arg so that it prints out the correct extras to install like torchgeo[models] vs torchgeo[datasets] then I'll have to go back and fix everywhere that it's used.

edit: @adamjstewart I went ahead and just changed the error message. PTAL

[Copilot]

Pull Request Overview

Adds the Delineate Anything YOLO segmentation models to TorchGeo by introducing a new yolo builder, registering its pretrained weights, and wiring up dependencies, tests, documentation, and CI.

• Introduce torchgeo/models/yolo.py with yolo function and YOLO_Weights
• Add a models extra for installing ultralytics and update CI/configs
• Expand tests, docs, torch hub integration, and project metadata to support YOLO

Reviewed Changes

Copilot reviewed 14 out of 14 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
torchgeo/models/yolo.py	New YOLO wrapper and pretrained weight definitions
torchgeo/models/api.py	Register yolo builder and YOLO_Weights in API
torchgeo/models/init.py	Export yolo and YOLO_Weights in public API
torchgeo/datasets/utils.py	Extend lazy_import instructions for models extra
tests/models/test_yolo.py	Add unit tests for yolo builder and weights
tests/models/test_api.py	Include YOLO in API tests with conditional skip
requirements/models.txt	Pin ultralytics for the models extra
pyproject.toml	Declare models extra and update all group
hubconf.py	Expose yolo in Torch Hub and add ultralytics
docs/api/models.rst	Document yolo function and YOLO_Weights
docs/api/weights/agnostic.csv	List YOLO weights metadata
.github/workflows/tests.yaml	Install models requirements in CI
.github/workflows/style.yaml	Install models requirements in style CI job

Comments suppressed due to low confidence (2)

tests/models/test_yolo.py:37

• Add an assertion to verify the returned object is an instance of nn.Module, e.g., model = yolo(...); assert isinstance(model, nn.Module).

        yolo(weights=None, model='yolo11n.yaml', task='segment')

tests/models/test_yolo.py:40

• Include an assertion to confirm that yolo(weights=mocked_weights) returns an nn.Module, for example: model = yolo(weights=mocked_weights); assert isinstance(model, nn.Module).

        yolo(weights=mocked_weights)