Handle missing detection/segmentation label files

[guarin]

What has changed and why?

• Add option to skip missing label files
• Treat missing label files by default as having no annotations

Follow up from #580

How has it been tested?

• 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)

[Copilot]

Pull request overview

Adds support for YOLO object detection and instance segmentation datasets to handle missing label files more gracefully by defaulting to “no annotations” and optionally skipping samples without labels.

Changes:

• Treat missing YOLO label files as empty annotations in both object detection and instance segmentation datasets.
• Add skip_if_label_file_missing option to optionally exclude samples without label files.
• Document the behavior change in CHANGELOG.md.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
src/lightly_train/_data/yolo_object_detection_dataset.py	Handle missing label files in __getitem__, add skip flag to args and listing logic.
src/lightly_train/_data/yolo_instance_segmentation_dataset.py	Same missing-label handling and skip flag wiring for instance segmentation.
CHANGELOG.md	Notes the default behavior change and how to configure skipping.

[guarin]

/review

[IgorSusmelj]

LGTM

Nit: I'd mention this in the docs. Common questions I'd have that I'd like to get answered:

• can I use images without annotations as negatives? E.g. supergradients and other frameworks do not allow that which in some cases is quite limiting. Sometimes I want to explicitly add images as negatives without any labels in them.
• (similar question): What happens with images that have no labels?