Improving Layout Representation Learning Across Inconsistently Annotated Datasets via Agentic Harmonization

2026-04-13

AI Summary

This work addresses the performance degradation in cross-dataset document layout detection caused by inconsistent annotation standards, particularly conflicts between semantic category definitions and bounding box granularity. To resolve this, the study introduces an agent-driven annotation harmonization mechanism—the first of its kind—that leverages a vision-language model to align both semantic labels and spatial boundaries across heterogeneous datasets prior to training. This unified representation achieves simultaneous semantic and geometric alignment, effectively restoring the underlying feature space structure. Evaluated on SCORE-Bench, the approach yields substantial improvements: a TEDS score of 0.814 for table structure recognition, a detection F-score of 0.883, and a reduced bounding box overlap error of 0.016, while producing more compact and discriminative embedding representations.

Technology Category

• Computer Vision: Language and Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Fine-tuning object detection (OD) models on combined datasets assumes annotation compatibility, yet datasets often encode conflicting spatial definitions for semantically equivalent categories. We propose an agentic label harmonization workflow that uses a vision-language model to reconcile both category semantics and bounding box granularity across heterogeneous sources before training. We evaluate on document layout detection as a challenging case study, where annotation standards vary widely across corpora. Without harmonization, naïve mixed-dataset fine-tuning degrades a pretrained RT-DETRv2 detector: on SCORE-Bench, which measures how accurately the full document conversion pipeline reproduces ground-truth structure, table TEDS drops from 0.800 to 0.750. Applied to two corpora whose 16 and 10 category taxonomies share only 8 direct correspondences, harmonization yields consistent gains across content fidelity, table structure, and spatial consistency: detection F-score improves from 0.860 to 0.883, table TEDS improves to 0.814, and mean bounding box overlap drops from 0.043 to 0.016. Representation analysis further shows that harmonized training produces more compact and separable post-decoder embeddings, confirming that annotation inconsistency distorts the learned feature space and that resolving it before training restores representation structure.

Problem

• annotation inconsistency
• layout representation learning
• object detection
• dataset harmonization
• bounding box granularity

Innovation

• agentic harmonization
• layout representation learning
• annotation inconsistency
• vision-language model
• object detection