ConRTF: Edge-Constrained Boundary Distribution Refinement for Realtime TransFormer Table Structure Recognition
2026-07-01 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial Intelligence
AI summaryⓘ
The authors work on improving how computers recognize the layout of tables in document images, focusing on accurately finding the boundaries between rows and columns. They point out that rows and columns have different roles and should be treated differently during training. To address this, they created a special training method called Edge-constrained Fine-grained Localization loss (EFL) that pays more attention to the right boundaries—horizontal for rows and vertical for columns. Their approach improves accuracy without slowing down the system and works well even with less training data. Tests show their method performs better than existing real-time detectors on several datasets.
Table Structure Recognitionboundary localizationrow and column layoutEdge-constrained Fine-grained Localization lossdistribution-based boundary refinementreal-time detectorPubTables-1M datasetRT-DETRv2YOLOv10-11GriTS metric
Authors
Eliott Thomas, Tri-Cong Pham, Mickael Coustaty, Aurelie Joseph, Gaspar Deloin, Vincent Poulain d'Andecy, Jean-Marc Ogier, Antoine Doucet
Abstract
Table Structure Recognition (TSR) aims to recover the row and column layout of tables from document images, a key step in document understanding pipelines. Accurate TSR depends on precise boundary localization: small errors in row or column boundaries can propagate into incorrect cell assignments and structural inconsistencies. Yet detection-based approaches treat table elements as generic objects, ignoring a fundamental property of table layout: rows and columns play structurally distinct roles and their boundaries carry unequal importance. We propose an Edge-constrained Fine-grained Localization loss (EFL) that formalizes this structural asymmetry by encoding table-specific geometric priors into the training objective: row-like elements are supervised with emphasis on their horizontal boundaries, while column-like elements prioritize vertical boundaries. Implemented within a real-time detector with distribution-based boundary refinement (D-FINE), EFL operates during training only and guides boundary refinement toward structurally meaningful adjustments with no change to the inference pipeline. The proposed approach, ConRTF, is also data-efficient, maintaining robust accuracy with as few as 2k--3k annotated tables. Experiments on PubTables-1M and two private datasets show consistent improvements over the optimized baseline and several real-time detectors including RT-DETRv2 and YOLOv10-11, with gains of up to +1.6 GriTS points at equal inference speed.