ml-pipes-ultralytics Index¶

This page catalogs the public ml_pipes.ultralytics package surface.

For the Ultralytics Model API comparison, see coverage.md. For the package overview and examples, see the repository README.

Public Modules¶

Module	Contents
`yolo`	`Predict`, `Embed`, and `Track` operators, plus the native source and configuration type aliases.
`results`	Operators wrapping configurable native `Results` methods.

Model Inference¶

See the upstream Ultralytics Model reference.

Operator	Input -> Output	Notes
`yolo.Predict(model="yolo26n.pt", task=None, verbose=False, **predict_options)`	`Sources` -> `list[Results]`	Runs non-streaming native YOLO prediction.
`yolo.Embed(model="yolo26n.pt", task=None, verbose=False, **embed_options)`	`Sources` -> `list[Tensor]`	Runs non-streaming native YOLO embedding. With no `embed` option, Ultralytics selects its default embedding layer.
`yolo.Track(model="yolo26n.pt", task=None, verbose=False, persist=False, **track_options)`	`Sources` -> `list[Results]`	Runs non-streaming native YOLO tracking. `persist` controls tracker state across calls to the same operator.

Result Operations¶

results is the module containing the result operators below. See the upstream Ultralytics Results reference.

Operator	Input -> Output	Notes
`results.To(...)`	`Results` -> `Results`	Moves native result tensors to a device or dtype.
`results.Plot(...)`	`Results` -> `ndarray \\| Image`	Renders an annotated image.
`results.Show(...)`	`Results` -> `Results`	Displays an annotated image and passes the result through.
`results.ToDataFrame(...)`	`Results` -> `polars.DataFrame`	Converts a result to a Polars dataframe.
`results.ToCSV(...)`	`Results` -> `str`	Converts a result to CSV.
`results.ToJSON(...)`	`Results` -> `str`	Converts a result to JSON.
`results.SaveTXT(...)`	`Results` -> `Results`	Writes labels and passes the result through.
`results.SaveCrop(...)`	`Results` -> `Results`	Writes detected-object crops and passes the result through.
`results.Save(...)`	`Results` -> `Results`	Saves an annotated image and passes the result through.
`results.Summary(...)`	`Results` -> `list[dict[str, Any]]`	Converts a result to summary dictionaries.

The argument-less Results methods cpu, numpy, cuda, new, and verbose can be used directly as unbound methods in a pipeline.

Inspection¶

Importing ml_pipes.ultralytics registers the BGR convention for bare np.ndarray image previews, matching Ultralytics and OpenCV. Inspection converts these previews to RGB without changing the source arrays. Other arrays retain their shape and dtype summaries. Native Results values show text summaries of their predictions.

For pipelines using RGB arrays, override the convention on your inspector:

import numpy as np
from ml_pipes.inspection import PipelineInspector, ndarray_image_formatter

inspector = PipelineInspector().register_value_formatter(
    np.ndarray, ndarray_image_formatter(default_color_space="RGB"), override=True,
)