Add Classification and Semantic Segmentation tasks for Ultralytics

[kprokofi]

Summary

How to test

Checklist

• The PR title and description are clear and descriptive
• I have manually tested the changes
• All changes are covered by automated tests
• All related issues are linked to this PR (if applicable)
• Documentation has been updated (if applicable)

[github-actions]

⏱️ Backend import time — app.main

Accelerator	Cumulative (s)	Self (s)
cpu	3.917	0.003
xpu	4.689	0.004
cuda	5.052	0.004

[github-actions]

🐳 Docker image sizes

Device	Size
cpu	4013.2 MB (3.92 GB)
xpu	10801.8 MB (10.55 GB)
cuda	10208.4 MB (9.97 GB)

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 52.94118% with 296 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
library/src/getitune/backend/ultralytics/engine.py	32.40%	73 Missing ⚠️
...e/backend/ultralytics/validators/classification.py	32.91%	53 Missing ⚠️
...y/src/getitune/backend/ultralytics/data/adapter.py	56.70%	42 Missing ⚠️
...une/backend/ultralytics/trainers/classification.py	43.10%	33 Missing ⚠️
...kend/ultralytics/trainers/semantic_segmentation.py	42.22%	26 Missing ⚠️
...y/src/getitune/backend/ultralytics/models/utils.py	36.36%	14 Missing ⚠️
...itune/backend/ultralytics/models/classification.py	76.36%	13 Missing ⚠️
...nd/ultralytics/validators/semantic_segmentation.py	48.00%	13 Missing ⚠️
...ackend/ultralytics/models/semantic_segmentation.py	75.00%	10 Missing ⚠️
...y/src/getitune/backend/ultralytics/data/collate.py	82.35%	9 Missing ⚠️
... and 6 more

📢 Thoughts on this report? Let us know!

[Copilot]

Pull request overview

This PR extends the library/ Ultralytics backend integration to support classification (multi-class + multi-label) and semantic segmentation, aligning training/validation/export paths with GetiTune’s DataModule-driven workflow.

Changes:

• Added Ultralytics trainers/validators/models for classification and semantic segmentation, plus torchmetrics-based evaluation for these tasks.
• Refactored the Ultralytics dataset adapter/validator plumbing from include_masks to a more explicit task_kind dispatch (detect, segment, classify, multilabel, semantic).
• Added new Ultralytics recipes for YOLO26 classification and semantic segmentation variants, and expanded unit tests for the new task support.

Reviewed changes

Copilot reviewed 43 out of 43 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
library/tests/unit/backend/ultralytics/test_models.py	Adds unit tests for the semantic segmentation model wrapper.
library/tests/unit/backend/ultralytics/test_exporter.py	Tests exporter metadata mapping for semantic segmentation.
library/tests/unit/backend/ultralytics/test_data_bridge.py	Adds adapter/collate tests for semantic segmentation and segment-mode semantic masks.
library/tests/unit/backend/ultralytics/test_configurator.py	Updates “unsupported task” expectations and validates semantic segmentation recipes load.
library/src/getitune/recipe/semantic_segmentation/yolo26_n_sem.yaml	Adds YOLO26-N semantic segmentation recipe.
library/src/getitune/recipe/semantic_segmentation/yolo26_s_sem.yaml	Adds YOLO26-S semantic segmentation recipe.
library/src/getitune/recipe/semantic_segmentation/yolo26_m_sem.yaml	Adds YOLO26-M semantic segmentation recipe.
library/src/getitune/recipe/semantic_segmentation/yolo26_l_sem.yaml	Adds YOLO26-L semantic segmentation recipe.
library/src/getitune/recipe/semantic_segmentation/yolo26_x_sem.yaml	Adds YOLO26-X semantic segmentation recipe.
library/src/getitune/recipe/classification/multi_class_cls/yolo26_n_cls.yaml	Adds YOLO26-N multi-class classification recipe.
library/src/getitune/recipe/classification/multi_class_cls/yolo26_s_cls.yaml	Adds YOLO26-S multi-class classification recipe.
library/src/getitune/recipe/classification/multi_class_cls/yolo26_m_cls.yaml	Adds YOLO26-M multi-class classification recipe.
library/src/getitune/recipe/classification/multi_class_cls/yolo26_l_cls.yaml	Adds YOLO26-L multi-class classification recipe.
library/src/getitune/recipe/classification/multi_class_cls/yolo26_x_cls.yaml	Adds YOLO26-X multi-class classification recipe.
library/src/getitune/recipe/classification/multi_label_cls/yolo26_n_cls.yaml	Adds YOLO26-N multi-label classification recipe.
library/src/getitune/recipe/classification/multi_label_cls/yolo26_s_cls.yaml	Adds YOLO26-S multi-label classification recipe.
library/src/getitune/recipe/classification/multi_label_cls/yolo26_m_cls.yaml	Adds YOLO26-M multi-label classification recipe.
library/src/getitune/recipe/classification/multi_label_cls/yolo26_l_cls.yaml	Adds YOLO26-L multi-label classification recipe.
library/src/getitune/recipe/classification/multi_label_cls/yolo26_x_cls.yaml	Adds YOLO26-X multi-label classification recipe.
library/src/getitune/backend/ultralytics/validators/base.py	Switches adapter dispatch from _include_masks to _task_kind.
library/src/getitune/backend/ultralytics/validators/detection.py	Sets _task_kind for detection validator.
library/src/getitune/backend/ultralytics/validators/instance_segmentation.py	Sets _task_kind for instance segmentation validator.
library/src/getitune/backend/ultralytics/validators/classification.py	Adds classification and multi-label classification validators.
library/src/getitune/backend/ultralytics/validators/semantic_segmentation.py	Adds semantic segmentation validator with mask-shape bookkeeping.
library/src/getitune/backend/ultralytics/validators/init.py	Exposes new validators in the package API.
library/src/getitune/backend/ultralytics/trainers/base.py	Switches adapter dispatch from _include_masks to _task_kind.
library/src/getitune/backend/ultralytics/trainers/detection.py	Sets _task_kind for detection trainer.
library/src/getitune/backend/ultralytics/trainers/instance_segmentation.py	Sets _task_kind for instance segmentation trainer.
library/src/getitune/backend/ultralytics/trainers/classification.py	Adds classification and multi-label classification trainers.
library/src/getitune/backend/ultralytics/trainers/semantic_segmentation.py	Adds semantic segmentation trainer.
library/src/getitune/backend/ultralytics/trainers/init.py	Exposes new trainers in the package API.
library/src/getitune/backend/ultralytics/models/classification.py	Adds Ultralytics model wrappers for multi-class and multi-label classification.
library/src/getitune/backend/ultralytics/models/criterion.py	Adds BCE loss + sigmoid inference head wiring for multi-label classification.
library/src/getitune/backend/ultralytics/models/semantic_segmentation.py	Adds Ultralytics semantic segmentation model wrapper.
library/src/getitune/backend/ultralytics/models/init.py	Exposes new model wrappers in the package API.
library/src/getitune/backend/ultralytics/data/adapter.py	Adds task-kind dispatch and per-task item builders, including semantic/classification support.
library/src/getitune/backend/ultralytics/data/collate.py	Adds collate functions for classification, multi-label, and semantic segmentation.
library/src/getitune/backend/ultralytics/data/init.py	Re-exports new collate utilities.
library/src/getitune/backend/ultralytics/exporter/exporter.py	Extends task metadata mapping to classification + semantic segmentation.
library/src/getitune/backend/ultralytics/engine.py	Adds classification/semantic torchmetrics evaluation paths and improves metric flattening.
library/src/getitune/backend/ultralytics/tools/utils.py	Registers recipe subdirectories and supported task types for the new tasks.
library/src/getitune/backend/ultralytics/README.md	Documents newly supported Ultralytics tasks/models (table needs follow-up tweaks).
library/src/getitune/backend/ultralytics/init.py	Exposes new models at backend package level.

Comments suppressed due to low confidence (1)

library/src/getitune/backend/ultralytics/engine.py:1092

• _format_torchmetrics_results() intends to skip auxiliary metric keys like classes/map_per_class, but the current check compares the full prefixed path (e.g. test/classes) against _skip_keys (e.g. classes), so the skip never triggers and the function relies on the non-scalar fallback/logging instead.

        Adds keys like ``val/precision/<ClassName>``, ``val/recall/<ClassName>``,
        ``val/map_50/<ClassName>``, ``val/map/<ClassName>`` to *metrics* in-place.
        """
        names = getattr(results, "names", None)

[github-actions]

📊 Test coverage report

Metric	Coverage
Lines	60.5%
Functions	57.0%
Branches	52.4%
Statements	60.1%

[Copilot]

Pull request overview

Copilot reviewed 69 out of 70 changed files in this pull request and generated 5 comments.

Comments suppressed due to low confidence (7)

library/tests/unit/backend/ultralytics/tools/test_configurator.py:107

• This path points to library/tests/unit/assets/..., which does not exist in this repo; as a result the test will always skip instead of exercising the DataModule path. library/tests/assets/... is at Path(__file__).resolve().parents[4] / "assets".
  library/tests/unit/backend/ultralytics/tools/test_configurator.py:771
• This path points to library/tests/unit/assets/..., which does not exist in this repo; as a result the test will always skip instead of exercising the DataModule path. library/tests/assets/... is at Path(__file__).resolve().parents[4] / "assets".
  library/tests/unit/backend/ultralytics/tools/test_configurator.py:792
• This path points to library/tests/unit/assets/..., which does not exist in this repo; as a result the test will always skip instead of exercising the DataModule path. library/tests/assets/... is at Path(__file__).resolve().parents[4] / "assets".
  library/tests/unit/backend/ultralytics/tools/test_configurator.py:812
• This path points to library/tests/unit/assets/..., which does not exist in this repo; as a result the test will always skip instead of exercising the DataModule path. library/tests/assets/... is at Path(__file__).resolve().parents[4] / "assets".
  library/tests/unit/backend/ultralytics/tools/test_configurator.py:829
• This path points to library/tests/unit/assets/..., which does not exist in this repo; as a result the test will always skip instead of exercising the DataModule path. library/tests/assets/... is at Path(__file__).resolve().parents[4] / "assets".
  library/tests/unit/backend/ultralytics/tools/test_configurator.py:922
• This path points to library/tests/unit/assets/..., which does not exist in this repo; as a result the test will always skip instead of exercising the DataModule path. library/tests/assets/... is at Path(__file__).resolve().parents[4] / "assets".
  library/tests/unit/backend/ultralytics/tools/test_configurator.py:952
• This path points to library/tests/unit/assets/..., which does not exist in this repo; as a result the test will always skip instead of exercising the DataModule path. library/tests/assets/... is at Path(__file__).resolve().parents[4] / "assets".

[Copilot]

Pull request overview

Copilot reviewed 75 out of 77 changed files in this pull request and generated 2 comments.

Comments suppressed due to low confidence (1)

library/tests/unit/backend/ultralytics/exporter/test_exporter.py:643

• This unit test uses task_type="semantic_segmentation", but the actual semantic-segmentation model wrapper exports with task_type="segmentation" (see UltralyticsSemanticSegModel._export_parameters). The test currently won't catch the real metadata mapping behavior for semantic segmentation.

[itallix]

pythonic way is to use property instead of setter method:

Suggested change

	def set_datamodule(self, datamodule: DataModule | None) -> None:
	"""Bind a DataModule instance, overriding any class-level default."""
	self._datamodule = datamodule
	@property
	def datamodule(self) -> DataModule | None:
	"""Get the bound DataModule instance."""
	return self._datamodule
	@datamodule.setter
	def datamodule(self, value: DataModule | None) -> None:
	"""Bind a DataModule instance, overriding any class-level default."""
	self._datamodule = value

[itallix]

Since getter methods have the same semantics, it can be simplified with the class-level registry:

_GETITEM_MAPPING: ClassVar[dict[str, Callable]] = {
        "classify": _getitem_classify,
        "multilabel": _getitem_multilabel,
        "segment": _getitem_segment,
        "semantic": _getitem_semantic,
        "detect": _getitem_detect,
    }

...

# 1. Look up the method (returns unbound method)
get_item_fn = self._GETITEM_MAPPING.get(self._task_kind)

# 2. If not found, raise the error
if get_item_fn is None:
    msg = f"Unknown task_kind: {self._task_kind}"
    raise ValueError(msg)

# 3. Call it, explicitly passing 'self'
return get_item_fn(self, sample, img)