Bounding Box Label Propagation for Re-Annotation of Document Layout Analysis Datasets

2026-06-16 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

Bounding Box Label Propagation (BBLP) is a new pseudo-labelling framework designed to reduce re-annotation efforts for object detection instances in document layout analysis datasets. This approach tackles the significant time and cost associated with continuously refining class annotations in growing datasets. BBLP integrates visual, textual, and positional embeddings from object detection samples to create a joint embedding, enabling plug-and-play label propagation on partially annotated datasets. Evaluated on the D4LA layout analysis dataset, BBLP achieved a mean Average Precision (mAP) of 54.0% using only 10% labelled data. This performance corresponds to 81.6% of the fully supervised benchmark, demonstrating its potential for generating high-quality class annotations and significantly reducing manual effort in real-world document processing applications.

Key takeaway

For Machine Learning Engineers managing evolving document layout analysis datasets, Bounding Box Label Propagation (BBLP) offers a compelling solution to reduce re-annotation costs. If you are struggling with time-consuming manual annotation, consider BBLP's semi-supervised approach. It can achieve 81.6% of fully supervised performance with only 10% labelled data, significantly streamlining your workflow and improving annotation efficiency. Implement BBLP to automate high-quality bounding box re-classification.

Key insights

BBLP uses multi-modal embeddings and label propagation to efficiently re-annotate object detection bounding boxes, significantly reducing manual effort.

Principles

• Semi-supervised learning reduces re-annotation costs.
• Multi-modal embeddings enhance object detection context.
• Label propagation extends annotations from small subsets.

Method

BBLP integrates visual, textual, and positional embeddings into a joint embedding. This embedding then facilitates Label Propagation on partially annotated object detection datasets in a plug-and-play manner.

In practice

• Re-annotate document layout analysis datasets.
• Generate high-quality bounding box class annotations.
• Reduce manual effort in document processing.

Topics

• Bounding Box Label Propagation
• Document Layout Analysis
• Object Detection
• Semi-supervised Learning
• Pseudo-labelling
• Multi-modal Embeddings

Best for: AI Engineer, Research Scientist, CTO, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Try this Auto dataset labelling tool! — Automated pixel-perfect polygon and bounding box generation accelerates image dataset annotation.
• Per-pixel bounding-box regression + DBSCAN for handwritten word detection - visual walkthrough of WordDetectorNet [P] — Per-pixel bounding box regression with DBSCAN offers an anchor-free, NMS-free approach for handwritten word detection.
• SEAL: Semantic-aware Single-image Sticker Personalization with a Large-scale Sticker-tag Dataset — SEAL improves single-image sticker personalization in diffusion models by mitigating overfitting and enhancing contextual control.
• Distilling Examples into Task Instructions: Enhanced In-Context Learning for Real-World B2B Conversations — Distilling ICL examples into structured instructions significantly improves performance and efficiency for complex B2B conversation classification.
• F\textsuperscript{2}LP-AP: Fast \& Flexible Label Propagation with Adaptive Propagation Kernel — F²LP-AP offers a training-free, adaptive label propagation method for diverse graph structures.
• Benchmarking Open-Source Layout Detection Models for Data Snapshot Extraction from Institutional Documents — Generic layout models fail to extract meaningful data snapshots from complex institutional documents.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.