feat: Added support for per-class confidence thresholds in layout detection.

[TeHelado]

threshold_by_class option exist is config but it's not actually being used. This adds support to fine-tune layout detection with threshold_by_class.

[Copilot]

Pull request overview

This PR implements support for per-class confidence thresholds in layout detection. The threshold_by_class configuration option existed but was not being utilized in the code. This feature allows fine-tuning of detection confidence thresholds on a per-class basis, enabling more granular control over which layout elements are detected.

Changes:

• Added _apply_per_class_threshold method to filter detections based on per-class confidence thresholds
• Modified the process method to use a lower pre-threshold when per-class thresholds are configured, then apply per-class filtering afterward
• Fixed incorrect comment in config.yaml (label 23 was incorrectly labeled as "text" instead of "vertical_text")
• Added documentation examples showing both numeric and string key formats for threshold_by_class

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
glmocr/layout/layout_detector.py	Implements per-class threshold filtering logic and integrates it into the detection pipeline
glmocr/config.yaml	Fixes mislabeled comment for label 23 and adds documentation examples for string-based class keys

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.