Add Image Classification Implementation

[guarin]

What has changed and why?

• Add multiclass image classification
• Add multilabel image classification

Adds support for fine-tuning any model supported by LightlyTrain for classification.

Will make a follow-up PR with docs and changelog updates. The current PR is for testing.

How has it been tested?

• Unit tests
• Manually

Did you update CHANGELOG.md?

• Yes
• Not needed (internal change)

Did you update the documentation?

• Yes
• Not needed (internal change without effects for user)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0194499b11

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

Adds an image-classification task to LightlyTrain, including both multiclass and multilabel fine-tuning support, along with dataset handling, transforms, task model, and end-to-end training tests.

Changes:

• Introduces a new image_classification task model + train loop (losses/metrics/exports) with multiclass and multilabel modes.
• Adds image-classification dataset parsing (folder + CSV) and a dedicated collate function.
• Extends the public training API (train_image_classification) and adds integration tests.

Reviewed changes

Copilot reviewed 23 out of 23 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
tests/helpers.py	Renames/extends dataset test helpers for multiclass and multilabel classification datasets.
tests/_data/test_image_classification_dataset.py	Updates dataset tests to use new multiclass/multilabel data-args types and helper naming.
tests/_commands/test_train_task.py	Adds end-to-end training tests for image classification (multiclass + multilabel).
src/lightly_train/types.py	Updates typed batch/item shapes for image classification labels representation.
src/lightly_train/_transforms/view_transform.py	Updates link comment related to RandomResizedCrop interpolation choice.
src/lightly_train/_transforms/image_classification_transform.py	Introduces an Albumentations-based transform pipeline for classification train/val.
src/lightly_train/_task_models/image_classification/transforms.py	Adds default transform-args presets for classification train/val.
src/lightly_train/_task_models/image_classification/train_model.py	Adds the train loop (optimizer/scheduler/loss/metrics) for classification.
src/lightly_train/_task_models/image_classification/task_model.py	Adds the classification task model (head, predict, ONNX/TRT export, checkpoint head handling).
src/lightly_train/_task_models/image_classification/init.py	Adds package marker for the new task model submodule.
src/lightly_train/_task_models/dinov3_eomt_semantic_segmentation/task_model.py	Improves backbone-weight load log message to include the path.
src/lightly_train/_task_models/dinov3_eomt_panoptic_segmentation/task_model.py	Improves backbone-weight load log message to include the path.
src/lightly_train/_task_models/dinov3_eomt_instance_segmentation/task_model.py	Improves backbone-weight load log message to include the path.
src/lightly_train/_task_models/dinov2_ltdetr_object_detection/task_model.py	Improves backbone-weight load log message to include the path.
src/lightly_train/_task_models/dinov2_linear_semantic_segmentation/task_model.py	Improves backbone-weight load log message to include the path.
src/lightly_train/_task_models/dinov2_eomt_semantic_segmentation/task_model.py	Improves backbone-weight load log message to include the path.
src/lightly_train/_task_checkpoint.py	Adds a resolve_auto hook to checkpoint-args base for task-specific defaults.
src/lightly_train/_models/package_helpers.py	Adds load_weights plumbing when constructing wrapped models.
src/lightly_train/_data/task_batch_collation.py	Adds an image-classification collate function producing ImageClassificationBatch.
src/lightly_train/_data/image_classification_dataset.py	Adds task type selection (multiclass vs multilabel) and CSV validation behavior.
src/lightly_train/_commands/train_task_helpers.py	Updates checkpoint-args validation to use per-task checkpoint args class + auto-resolution.
src/lightly_train/_commands/train_task.py	Adds train_image_classification entry point and task config types.
src/lightly_train/init.py	Exposes train_image_classification in the public package API.