A curated research map for understanding tables as visual and multimodal objects.
Table perception Β· structure recognition Β· multimodal reasoning Β· complex tables Β· synthetic training data
Updated 2026-08-13. 175 papers Β· 18 practical generation projects.
- π 1. Visual Table Perception & Recognition
- π§ 2. Multimodal Table Understanding & Reasoning
- π 3. Multimodal Table Data Generation
- π§ͺ 4. Benchmarks & Datasets
- π 5. Evaluation & Analysis
- π 6. Surveys
- π Related Work
Methods that locate tables, recover topology, recognize cell content, or reconstruct structured markup from pixels. 72 papers.
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2025 | RAPTOR: Refined Approach for Product Table Object Recognition | WACV Workshops | Detects and organizes visual objects inside product comparison tables. | β |
| 2024 | TabSniper: Towards Accurate Table Detection and Structure Recognition for Bank Statements | CODS-COMAD | Targets long, varied transaction tables and introduces BankTabNet for bank-statement extraction. | β |
| 2024 | TC-OCR: TableCraft OCR for Efficient Detection and Recognition of Table Structure and Content | arXiv | Combines table localization, structural parsing, and content recognition in an efficient pipeline. | β |
| 2022 | Visual Understanding of Complex Table Structures from Document Images | WACV | Introduces TOD-Net and the TUCD benchmark for complex bordered and borderless tables. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | InstructTable: Improving Table Structure Recognition Through Instructions | arXiv | Uses instruction pretraining and introduces BCDSTab for balanced complex synthetic table evaluation. | β |
| 2026 | TRivia: Self-supervised Fine-tuning of Vision-Language Models for Table Recognition | CVPR | Learns from unlabeled table images through consistency sampling and a verifiable QA proxy task. | π» Code |
| 2025 | A One-Stage Network for Table Structure Recognition | arXiv | Predicts table structure in a single-stage architecture designed to simplify the recognition pipeline. | β |
| 2025 | Enhanced Table Structure Recognition with Multi-Modal Approach | WASP | Fuses visual and textual streams with separate structure and cell-box decoders. | β |
| 2025 | Enhancing Table Recognition with Vision LLMs: Benchmark and Neighbor-Guided Toolchain Reasoner | IJCAI | Benchmarks vision LLM recognition and uses neighboring cells plus tools to repair structural predictions. | π» Code |
| 2025 | From Pixels to Tables: Reconstructing Complex Tables from Document Images | IJDAR | Reconstructs complex tabular topology from document pixels with a structure-aware recognition pipeline. | β |
| 2025 | LORE++: Logical Location Regression Network for Table Structure Recognition with Pre-training | Pattern Recognition | Extends logical-location regression with table-focused pre-training for stronger structural generalization. | β |
| 2025 | SepFormer: Coarse-to-Fine Separator Regression for Table Structure Recognition | arXiv | Regresses table separators in coarse-to-fine stages for robust structure recovery. | β |
| 2025 | SLANet-1M: A Lightweight and Efficient Model for Table Recognition with Minimal Computational Cost | SwissText | A roughly one-million-parameter table recognizer designed for fast, CPU-friendly inference. | β |
| 2025 | TableStructureFormer: An Improved Masked-Attention Mask Transformer Model with Long-Distance Feature Aggregation and Deep Detail Supervision for Table Structure Recognition | Complex & Intelligent Systems | Combines masked attention, long-range aggregation, and detail supervision for row-column segmentation. | β |
| 2025 | TABLET: Table Structure Recognition Using Encoder-only Transformers | ICDAR | Formulates structure recognition with an encoder-only transformer rather than an autoregressive decoder. | β |
| 2025 | Treading Towards Privacy-Preserving Table Structure Recognition | WACV | Studies structure recovery when sensitive table content must be protected. | β |
| 2025 | TRH2TQA: Table Recognition with Hierarchical Relationships to Table Question-Answering on Business Documents | WACV | Connects hierarchical table recognition directly to business-document QA. | β |
| 2024 | ClusterTabNet: Supervised Clustering Method for Table Detection and Table Structure Recognition | ICDAR | Casts table and cell grouping as supervised clustering for joint detection and structure recovery. | β |
| 2024 | DTSM: Toward Dense Table Structure Recognition with Text Query Encoder and Adjacent Feature Aggregator | ICDAR | Uses text-conditioned queries and adjacent-cell aggregation for dense table layouts. | β |
| 2024 | End to End Table Transformer | ICDAR | Adapts DETR queries and training for single-model table extraction from document images. | β |
| 2024 | Enhancing Transformer-Based Table Structure Recognition for Long Tables | PRCV | Adapts Transformer recognition to extreme aspect ratios and long table sequences. | β |
| 2024 | LRATNet: Local-Relationship-Aware Transformer Network for Table Structure Recognition | MMM | Connects local row, column, and spatial relationships with global Transformer features. | β |
| 2024 | Multi-Modal Attention Based on 2D Structured Sequence for Table Recognition | PRCV | Jointly models structure-language sequences and cell boxes with a table-specific 2D attention layer. | β |
| 2024 | RTSR: A Real-Time Table Structure Recognition Approach | ECAI | Targets low-latency structure recognition while retaining accuracy on complex table layouts. | β |
| 2024 | Self-Supervised Pre-Training for Table Structure Recognition Transformer | AAAI SDU Workshop | Shows self-supervised visual pre-training can close the gap between linear-patch and CNN-based TSR encoders. | π» Code |
| 2024 | SEMv2: Table Separation Line Detection Based on Instance Segmentation | Pattern Recognition | Detects separation lines as instances and introduces iFLYTAB for complex and deformed tables. | π» Code |
| 2024 | SEMv3: A Fast and Robust Approach to Table Separation Line Detection | IJCAI | Regresses offsets from keypoint proposals to recover separators in wireless and deformed tables. | π» Code |
| 2024 | SPRINT: Script-Agnostic Structure Recognition in Tables | ICDAR | Separates structure cues from script-specific text to improve multilingual table parsing. | π» Code |
| 2024 | TableStrRec: Framework for Table Structure Recognition in Data Sheet Images | IJDAR | Targets irregular structures and heterogeneous layouts in technical data-sheet imagery. | β |
| 2024 | TFLOP: Table Structure Recognition Framework with Layout Pointer Mechanism | IJCAI | Uses layout-aware pointer prediction to recover table structure. | β |
| 2023 | Formerge: Recover Spanning Cells in Complex Table Structure Using Transformer Network | ICDAR | Focuses on recovering row- and column-spanning cells after initial grid prediction. | β |
| 2023 | GrabTab: Grab What You Need at Once for Table Recognition | IJCAI | Models table components and their relations jointly to avoid multi-stage error propagation. | β |
| 2023 | High-Performance Transformers for Table Structure Recognition Need Better Relative Positional Encodings | arXiv | Shows that table-aware relative positions materially improve transformer-based structure recognition. | β |
| 2023 | Improving Table Structure Recognition with Visual-Alignment Sequential Coordinate Modeling | CVPR | Generates markup and aligned cell coordinates to couple sequence decoding with visual location. | β |
| 2023 | LORE: Logical Location Regression Network for Table Structure Recognition | AAAI | Regresses logical row and column coordinates and refines them with neighboring-cell constraints. | β |
| 2023 | Optimized Table Tokenization for Table Structure Recognition | ICDAR | Introduces OTSL, a compact token language designed to simplify valid structure generation. | β |
| 2023 | PyramidTabNet: Transformer-Based Table Recognition in Image-Based Documents | ICDAR | Uses a convolution-free pyramid vision transformer plus generative augmentation for document tables. | π» Code |
| 2023 | Rethinking Detection Based Table Structure Recognition for Visually Rich Document Images | arXiv | Revisits detection labels and post-processing for visually rich, irregular document tables. | β |
| 2023 | Rethinking Image-Based Table Recognition Using Weakly Supervised Methods | ICPRAM | Reduces dependence on fully annotated table images through weak supervision. | β |
| 2023 | Robust Table Structure Recognition with Dynamic Queries Enhanced Detection Transformer | Pattern Recognition | Introduces dynamically initialized and enhanced queries for detection-based structure recognition. | β |
| 2023 | Scene Table Structure Recognition with Segmentation and Key Point Collaboration | ICDAR | Fuses cell segmentation and key points for geometrically distorted scene tables and online data generation. | β |
| 2023 | Scene Table Structure Recognition with Segmentation Collaboration and Alignment | Pattern Recognition Letters | Collaborates and aligns segmentation branches to recover structures from scene-captured tables. | β |
| 2023 | TableVLM: Multi-modal Pre-training for Table Structure Recognition | ACL | Pretrains on aligned visual and textual signals and introduces the million-scale ComplexTable resource. | β |
| 2022 | Complex Table Structure Recognition in the Wild Using Transformer and Identity Matrix-Based Augmentation | ICFHR | Targets camera-based tables and introduces identity-matrix-aware TabSplitter augmentation. | β |
| 2022 | Neural Collaborative Graph Machines for Table Structure Recognition | CVPR | Models cell relations with collaborative graph reasoning for complex structures. | β |
| 2022 | Split, Embed and Merge: An Accurate Table Structure Recognizer | Pattern Recognition | Introduces the split-embed-merge paradigm and fuses visual and semantic features at the grid level. | π» Code |
| 2022 | TableFormer: Table Structure Understanding with Transformers | CVPR | Generates table structure and regresses cell boxes while leveraging PDF text tokens when available. | β |
| 2022 | TSRFormer: Table Structure Recognition with Transformers | ACM Multimedia | Detects table separators with transformers and merges grid cells into spanning structures. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2024 | Multi-Cell Decoder and Mutual Learning for Table Structure and Character Recognition | ICDAR | MuTabNet couples multi-cell structural decoding with character recognition through mutual learning. | β |
| 2024 | UniTabNet: Bridging Vision and Language Models for Enhanced Table Structure Understanding | Findings of EMNLP | Aligns visual and linguistic table representations for structure and content recognition. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | CSPO: Alleviating Reward Ambiguity for Structured Table-to-LaTeX Generation | ACL | Optimizes table-to-LaTeX generation with structure-sensitive policy rewards. | β |
| 2026 | Tables Decoded: DELTA for Structure, TARQA for Understanding | WACV | Pairs a modular table extractor using OTSL with a downstream table QA system. | β |
| 2026 | TableSeq: Unified Generation of Structure, Content, and Layout | arXiv | Generates structure, cell content, and layout within a unified sequence formulation. | π» Code |
| 2026 | TDATR: Improving End-to-End Table Recognition via Table Detail-Aware Learning and Curriculum Learning | CVPR | Targets fine-grained table details and staged learning for end-to-end structure and content recovery. | β |
| 2025 | DREAM: Document Reconstruction via End-to-end Autoregressive Model | ACM Multimedia | Autoregressively reconstructs document structure and content, including complex tables, in one parsing model. | β |
| 2025 | Table2LaTeX-RL: High-Fidelity LaTeX Code Generation from Table Images via Reinforced Multimodal Language Models | NeurIPS | Applies reinforcement learning to structure- and content-faithful table image-to-LaTeX generation. | β |
| 2024 | Enhancing Table Structure Recognition via Bounding Box Guidance | ICPR | BGTR guides HTML generation with predicted cell boxes and introduces the synthetic SNSTab scene dataset. | β |
| 2024 | OmniParser: A Unified Framework for Text Spotting, Key Information Extraction and Table Recognition | CVPR | Uses one prompt-conditioned architecture for three document parsing tasks, including table recognition. | β |
| 2024 | RCAM-Transformer: A Novel Approach to Table Reconstruction Using Row-Column Attention Mechanism | DAS | Builds row-column attention and reconstruction logic around the distinct geometry of table axes. | β |
| 2024 | TableRocket: An Efficient and Effective Framework for Table Reconstruction | PRCV | Provides an efficiency-oriented pipeline for reconstructing complete tables from cell-level predictions. | β |
| 2024 | UniTable: Towards a Unified Framework for Table Recognition via Self-Supervised Pretraining | arXiv | Unifies structure, content, and cell-box extraction under a pixel-to-language-modeling objective. | π» Code |
| 2023 | An End-to-End Local Attention Based Model for Table Recognition | ICDAR | Uses local attention for end-to-end table recognition and contributes the PubTabNet250 evaluation set. | β |
| 2023 | An End-to-End Multi-Task Learning Model for Image-Based Table Recognition | VISIGRAPP | Jointly learns structure and content prediction for image-based table recognition. | β |
| 2023 | Divide Rows and Conquer Cells: Towards Structure Recognition for Large Tables | IJCAI | Decomposes large tables row-wise before cell reconstruction to handle dense and oversized structures. | β |
| 2023 | GridFormer: Towards Accurate Table Structure Recognition via Grid Prediction | ACM Multimedia | Predicts row-column grids and spanning-cell relationships for table reconstruction. | β |
| 2023 | Tables to LaTeX: Structure and Content Extraction from Scientific Tables | IJDAR | Extracts both structure and cell content from scientific table images into LaTeX. | β |
| 2023 | TRACE: Table Reconstruction Aligned to Corner and Edges | ICDAR | Aligns predicted table corners and edges to reconstruct cells and spanning structures. | β |
| 2022 | A Structure-Focused Deep Learning Approach for Table Recognition from Document Images | COMPSAC | Prioritizes structural reconstruction when extracting tables from heterogeneous document imagery. | β |
| 2022 | GTRNet: A Graph-Based Table Reconstructed Network | ICTAI | Groups detected textual components through a graph network to reconstruct table topology. | β |
| 2022 | Table Structure Recognition and Form Parsing by End-to-End Object Detection and Relation Parsing | Pattern Recognition | Jointly optimizes region detection and graph relation parsing for tables and template-free forms. | β |
| 2022 | Table Structure Recognition with Conditional Attention | arXiv | Conditions sequence decoding on visual structure cues for image-to-markup recognition. | β |
| 2022 | TRUST: An Accurate and End-to-End Table Structure Recognizer Using Splitting-Based Transformers | arXiv | Uses learned row and column splitting followed by span reconstruction. | β |
Visual and multimodal methods for answering, calculating, retrieving, grounding, and reasoning over standard and complex tables. 21 papers.
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | Multimodal Table Understanding with Difficulty-aware Reinforcement Learning | AAAI | Introduces MM-Table-R1 with task- and data-level curricula based on table complexity. | β |
| 2026 | V-TableR1: Process-Supervised Multimodal Table Reasoning with Critic-Guided Policy Optimization | arXiv | Uses a critic VLM and process-guided policy optimization to supervise intermediate visual-table reasoning. | β |
| 2025 | Can GRPO Boost Complex Multimodal Table Understanding? | EMNLP | Table-R1 combines warm-up, TEDS-based perception alignment, and hint-completion GRPO. | β |
| 2024 | PixT3: Pixel-based Table-to-Text Generation | ACL | Generates faithful descriptions directly from table pixels without relying on a serialized table input. | β |
| 2024 | TabPedia: Towards Comprehensive Visual Table Understanding with Concept Synergy | NeurIPS | Couples low-level table recognition with high-level concept reasoning and introduces ComTQA. | π» Code |
| 2022 | End-to-End Compound Table Understanding with Multi-Modal Modeling | ACM Multimedia | Introduces the bilingual ComFinTab benchmark and CTUNet for compound table recognition and relation understanding. | π» Code |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | CoReTab: Improving Multimodal Table Understanding with Code-driven Reasoning | EACL | Builds verified code-grounded traces and combines pretraining, supervised tuning, and GRPO. | β |
| 2025 | ExpliCIT-QA: Explainable Code-Based Image Table Question Answering | arXiv | Produces executable code explanations for arithmetic and logical QA over table images. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | TableDART: Dynamic Adaptive Multi-Modal Routing for Table Understanding | ICLR 2026 | Routes between visual and serialized table representations according to example difficulty. | π» Code |
| 2026 | TABQAWORLD: Optimizing Multimodal Reasoning for Multi-Turn Table Question Answering | arXiv | Dynamically selects visual or textual table state representations during multi-turn reasoning. | β |
| 2026 | Thinking with Tables: Enhancing Multi-Modal Tabular Understanding via Neuro-Symbolic Reasoning | arXiv | Combines visual table perception with symbolic operations for interpretable tabular reasoning. | β |
| 2025 | Enhancing Large Vision-Language Models with Layout Modality for Table Question Answering on Japanese Annual Securities Reports | IIAI AAI | Adds OCR text and layout features to LVLMs for financial table cell QA in Japanese filings. | β |
| 2025 | HIPPO: Enhancing the Table Understanding Capability of LLMs through Hybrid-Modal Preference Optimization | arXiv | Aligns preferences across table images and textual table representations. | β |
| 2025 | Multimodal Tabular Reasoning with Privileged Structured Information | NeurIPS | Uses table markup only as privileged training supervision while retaining image-only inference. | β |
| 2025 | TableMoE: Neuro-Symbolic Routing for Structured Expert Reasoning in Multimodal Table Understanding | arXiv | Routes table questions between neural perception and symbolic execution components. | β |
| 2024 | Multimodal Table Understanding | ACL | Defines a broad 24-task setting and introduces MMTab plus Table-LLaVA for recognition and reasoning. | π» Code Β· ποΈ Data |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | TabRAG: Efficient Table Retrieval and Understanding with Multimodal Large Language Models | arXiv | Unifies visual table reranking and answer generation for retrieval-augmented table understanding. | β |
| 2025 | RITT: A Retrieval-Assisted Framework with Image and Text Table Representations for Table Question Answering | TRL Workshop | Retrieves a relevant subtable and dynamically selects visual or textual table representations by question type. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | Decoupling Skeleton and Flesh: Efficient Multimodal Table Reasoning with Disentangled Alignment and Structure-aware Guidance | arXiv | Separates structural abstraction from semantic grounding and adds structure-aware inference guidance. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2024 | Knowledge-Aware Reasoning over Multimodal Semi-Structured Tables | Findings of EMNLP | Introduces MMTABQA, where semi-structured tables interleave text cells with image cells. | β |
| 2023 | M-TBQA: Multimodal Table-Based Question Answering | ICMLCA | An early task and model for question answering over tables whose cells include both text and images. | β |
Pipelines and tools for creating table contents, schemas, rendered pixels, instructions, and validated synthetic training data. 21 papers.
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2025 | CTSyn: A Foundation Model for Cross Tabular Data Generation | ICLR | Pretrains a cross-table foundation model for transferable structured-value generation across heterogeneous tables. | β |
| 2024 | Mixed-Type Tabular Data Synthesis with Score-based Diffusion in Latent Space | ICLR | Generates mixed categorical and numerical table contents with score-based diffusion in a learned latent space. | π» Code |
| 2023 | Language Models are Realistic Tabular Data Generators | ICLR | GReaT serializes structured rows as natural language and fine-tunes language models to generate realistic table contents. | π» Code |
| 2023 | REaLTabFormer: Generating Realistic Relational and Tabular Data using Transformers | arXiv | Generates single-table and relational structured values while preserving dependencies useful for downstream rendering. | π» Code |
| 2023 | TabuLa: Harnessing Language Models for Tabular Data Synthesis | arXiv | Adapts language models for efficient synthetic table-content generation with compact table-specific tokenization. | π» Code |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | TableNet: A Large-Scale Table Dataset with LLM-Powered Autonomous Generation | arXiv | Uses cooperating agents for schema planning, layout construction, and content filling before rendering HTML tables with exact structure labels. | π» Code |
| 2025 | Map&Make: Generating & Evaluating Scientific Text-to-Table Data | ACL | Maps scientific text into atomic propositions and a latent schema before generating grounded tables. | π» Code Β· π Project |
| 2024 | Text-Tuple-Table: Towards Information Integration in Text-to-Table Generation via Global Tuple Extraction | EMNLP | Uses global tuple extraction as an intermediate representation for generating faithful tables from collections of text. | π» Code |
| 2022 | Text-to-Table: A New Way of Information Extraction | ACL | Frames information extraction as generating complex structured tables from text with structure-aware constraints. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2025 | FlexDoc: Parameterized Sampling for Diverse Multilingual Synthetic Documents for Training Document Understanding Models | EMNLP Industry Track | Samples multilingual semi-structured documents from stochastic schemas with parameterized layouts and table-boundary annotations. | β |
| 2025 | Scaling Text-Rich Image Understanding via Code-Guided Synthetic Multimodal Data Generation | ACL | Uses generated Python, HTML, and LaTeX programs to render text-rich images and build 2.7M instruction examples over 400K synthetic images. | π» Code Β· ποΈ Data Β· π Project |
| 2024 | ΞΌgat: Multimodal Generalized Academic Translator | ECCV Workshop | Includes a reproducible Markdown and LaTeX table generator that renders paired PDF or PNG pages with source markup. | π» Code |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2024 | Latent Diffusion for Guided Document Table Generation | ICDAR | Uses layout-guided latent diffusion to synthesize document tables for downstream recognition. | β |
| 2024 | Synthesizing Realistic Data for Table Recognition | ICDAR | Generates realistic financial table images to enlarge recognition training data without manual labeling. | π» Code |
| 2023 | Diffusion Models for Document Image Generation | ICDAR | Studies unconditional and layout-controlled document image synthesis, providing a direct pixel-generation route for table-bearing pages. | β |
| 2023 | Table and Image Generation for Investigating Knowledge of Entities in Pre-trained Vision and Language Models | ACL | Builds paired table and image generation data from roughly 200K Wikipedia infoboxes for entity-knowledge studies. | π» Code |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2025 | SynTab-LLaVA: Enhancing Multimodal Table Understanding with Decoupled Synthesis | CVPR | Decouples table rendering from QA synthesis to build SynTab at large scale with lower generation cost. | π» Code |
| 2025 | TableDreamer: Synthesizing High-Quality Table Instruction Tuning Data with Pluggable LLMs | Findings of ACL | Synthesizes table titles, contents, instructions, and responses with weakness-guided iterative exploration, yielding 27K instruction examples. | π» Code |
| 2025 | TabularMath: Understanding Math Reasoning over Tables with Large Language Models | arXiv | AutoT2T converts mathematical word problems into verified table-reasoning tasks and controllably perturbed variants. | π» Code |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2025 | GRAFT: Generator of Realistic And Formatted Tables | arXiv | Programmatically renders controlled table families and generates executable questions with machine-verifiable JSON or YAML answers. | β |
| 2025 | Why LLMs Are Bad at Synthetic Table Generation (and what to do about it) | arXiv | Diagnoses permutation sensitivity in LLM table synthesis and proposes permutation-aided fine-tuning to better preserve functional dependencies. | π» Code |
Projects are grouped under the same production stages as the papers above, so the literature and runnable resources stay in one place.
| Project | Produces | Best used for |
|---|---|---|
| TableDreamer official paper code |
Table titles, contents, instructions, and responses | Weakness-guided table instruction synthesis with pluggable LLMs |
| Map&Make official paper code |
Grounded scientific tables from unstructured text | Text-to-table data with proposition and schema intermediates |
| Synthetic Data Vault production toolkit |
Synthetic single-table, multi-table, and sequential values | Producing statistically coherent cell contents before rendering |
| GReaT official paper code |
Language-model-generated structured rows | Domain-adapted table content generation from small seed datasets |
| REaLTabFormer official paper code |
Synthetic relational and single-table values | Preserving dependencies across linked tables before visual rendering |
| TabSyn official paper code |
Diffusion-generated mixed-type table rows | High-fidelity numerical and categorical content synthesis |
| SDGX production toolkit |
Pluggable synthetic structured datasets | Composing privacy, quality, and generator modules in production pipelines |
| Project | Produces | Best used for |
|---|---|---|
| TableGeneration official paper code |
HTML tables, rendered images, and exact TSR labels | Autonomous schema, layout, and content generation with reproducible supervision |
| ΞΌgat dataset generator official paper code |
Markdown/LaTeX source paired with PDF and PNG pages | Academic-style tables with controlled templates and source markup |
| Great Tables production toolkit |
Styled HTML tables and exportable table images | Programmatic DataFrame-to-pixel rendering with rich styling |
| DoGe Document Generator community project |
Synthetic document pages with tables, text, and bounding boxes | Generating table-bearing page layouts and detection annotations |
| Synthetic OCR Image Generator community project |
Table images with Markdown and JSON ground truth | Quick OCR and TSR corpus generation from configurable templates |
| Project | Produces | Best used for |
|---|---|---|
| WikiTIG official paper code |
Entity-conditioned table and image generation pairs | Bootstrapping generation data from Wikipedia infoboxes |
| Project | Produces | Best used for |
|---|---|---|
| CoSyn / pixmo-docs official paper code |
Python/HTML/LaTeX renders plus QA and reasoning instructions | Scaling code-guided text-rich multimodal instruction data |
| SynTab-LLaVA official paper code |
Rendered table images and synthesized visual-table QA | Decoupled, large-scale multimodal table instruction construction |
| AutoT2T official paper code |
Verified mathematical table tasks and controlled perturbations | Generating executable numerical reasoning examples |
| Project | Produces | Best used for |
|---|---|---|
| Realistic Table Recognition Synthesizer official paper code |
Realistic financial table images and recognition labels | Draw-and-paste synthesis that retains complex real structures |
| TableVQA-Bench official paper code |
Controlled table visualizations and visual QA evaluations | Testing whether synthetic styles preserve content and reasoning difficulty |
Datasets and benchmark suites for recognition, understanding, robustness, and multimodal table settings. 26 papers.
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | UniTabBank: A Large-Scale Multi-Lingual, Multi-Layout, Multi-Type, Multi-Format Dataset for Table Detection | WACV | Broadens table detection coverage across languages, layouts, table types, and file formats. | β |
| 2025 | CISOL: An Open and Extensible Dataset for Table Structure Recognition in the Construction Industry | WACV | Provides an extensible recognition dataset designed for broader table layout diversity. | β |
| 2023 | WEATHERGOV+: A Table Recognition and Summarization Dataset to Bridge the Gap Between Document Image Analysis and Natural Language Generation | DocEng | Connects document-image table recognition with end-to-end natural-language table summarization. | β |
| 2022 | PubTables-1M: Towards Comprehensive Table Extraction From Unstructured Documents | CVPR | A million-table dataset with detailed structure annotations and a DETR-based Table Transformer baseline. | π» Code Β· ποΈ Data |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | INDOTABVQA: A Benchmark for Cross-Lingual Table Understanding in Bahasa Indonesia Documents | Findings of ACL | Contains 1,593 Indonesian document images and question sets in Indonesian, English, Hindi, and Arabic. | ποΈ Data |
| 2026 | MMFCTUB: Multi-Modal Financial Credit Table Understanding Benchmark | arXiv | A domain benchmark for structure perception, financial knowledge, and calculation across credit tables. | β |
| 2026 | TABVERSE: Benchmarking Cross-Format Table Understanding in LLMs and VLMs | arXiv | Aligns identical tables across HTML, Markdown, LaTeX, and rendered images to isolate representation effects. | π» Code Β· ποΈ Data |
| 2026 | ViTaB-A: Evaluating Multimodal Large Language Models on Visual Table Attribution | arXiv | Tests answer attribution across image, Markdown, and JSON table representations. | β |
| 2025 | Does Table Source Matter? Benchmarking and Improving Multimodal Scientific Table Understanding and Reasoning | OpenReview | Introduces MMSci and measures how scientific source domains affect visual-table reasoning. | β |
| 2025 | EviFiVQA: A Benchmark for Evidence-Grounded Multi-hop Reasoning in Financial VQA | ICDAR | Requires numerical multi-hop reasoning, evidence localization, and hierarchical aggregation over irregular financial tables. | β |
| 2025 | MTabVQA: Evaluating Multi-Tabular Reasoning of Language Models in Visual Space | arXiv | Evaluates reasoning that requires jointly reading multiple table images. | β |
| 2025 | TabComp: A Dataset for Visual Table Reading Comprehension | Findings of NAACL | Targets reading comprehension over visual tables with questions tied to table structure and content. | β |
| 2025 | Visual-TableQA: Open-Domain Benchmark for Reasoning over Table Images | arXiv | Generates open-domain LaTeX-rendered tables, reasoning-heavy questions, rationales, and multi-model verification. | β |
| 2024 | TableVQA-Bench: A Visual Question Answering Benchmark on Multiple Table Domains | arXiv | Renders existing table datasets into images and supplies 1,500 generated QA pairs across domains. | π» Code |
| 2024 | WikiDT: Visual-Based Table Recognition and Question Answering Dataset | ICDAR | Contains 70,919 images and 159,905 tables with hierarchical labels supporting both QA and recognition diagnosis. | ποΈ Data Β· π Project |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | A Benchmark and Method for Photographed Table Reasoning | Pattern Recognition | Introduces PCTR-16K and structure-aware chain-of-thought for Chinese tables photographed under real conditions. | π» Code |
| 2026 | TABLET: A Large-Scale Dataset for Robust Visual Table Understanding | ICLR | Aggregates more than four million examples over 21 tasks while preserving mostly original table visualizations. | π» Code |
| 2026 | WildTableBench: Benchmarking Multimodal Foundation Models on Table Understanding In the Wild | arXiv | A manually verified benchmark built from naturally occurring table images rather than template renders. | π» Code Β· ποΈ Data |
| 2025 | RealHiTBench: A Comprehensive Realistic Hierarchical Table Benchmark | arXiv | Focuses on realistic hierarchical tables and compares image and text presentation formats. | β |
| 2025 | TableEval: A Real-World Benchmark for Complex, Multilingual, and Multi-Structured Table Question Answering | EMNLP | Tests multilingual QA over complex real-world tables in both visual and serialized formats. | β |
| 2023 | A Large-Scale Dataset for End-to-End Table Recognition in the Wild | Scientific Data | Introduces TabRecSet with 38.1K English and Chinese tables from scanned and camera-captured scenes. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | MMTableBench: A Multi-level Multimodal Benchmark for Reasoning and Layout Complexity in Table QA | The Web Conference | Separates layout complexity from reasoning complexity for fine-grained diagnosis of multimodal TableQA. | π» Code Β· ποΈ Data |
| 2026 | MMTabReal: A Real-World Benchmark for Multimodal Table Understanding | Findings of ACL | Provides 500 real-world tables with 4,021 questions spanning embedded graphics, color encodings, and complex layouts. Formerly MMTBench. | β |
| 2026 | TableVista: Benchmarking Multimodal Table Reasoning under Visual and Structural Complexity | Findings of ACL | Expands 3,000 reasoning problems into 30,000 controlled visual variants to isolate style, structure, and vision-only failure modes. | β |
| 2025 | 2Columns1Row: A Benchmark for Textual and Multimodal Table Reasoning in Russian | Findings of EMNLP | Adds a Russian-language comparison of textual and visual table reasoning. | β |
| 2025 | Benchmarking Multimodal Large Language Models on Recognition and Understanding over Chemical Tables | arXiv | Evaluates chemical table recognition and reasoning where cells may contain domain-specific molecular graphics. | β |
Metrics, controlled evaluations, representation studies, and analyses of robustness and generalization. 11 papers.
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2025 | Benchmarking Table Extraction: Multimodal LLMs vs Traditional OCR | XLLM Workshop | Compares multimodal LLM extraction with Table Transformer plus OCR under structure- and content-aware metrics. | β |
| 2024 | Uncertainty Quantification in Table Structure Recognition | IRI | Studies predictive uncertainty so recognition systems can surface unreliable table structures. | β |
| 2023 | GriTS: Grid Table Similarity Metric for Table Structure Recognition | ICDAR | Defines factored grid-based metrics for topology, location, and content similarity. | β |
| 2022 | Evaluating Table Structure Recognition: A New Perspective | DAS | Analyzes weaknesses of existing evaluation and proposes geometry-aware TEDS-IOU. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | Zero-Shot Table Extraction in Business Documents: A Unified Benchmark with Realistic Evaluation | WACV | Evaluates table detection and structure recognition under a unified zero-shot business-document protocol. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2026 | Understanding and Bridging the Modality Gap When Text Is Rendered as Images | arXiv | Diagnoses why models underperform when equivalent textual content is supplied through pixels, including table benchmarks. | β |
| 2025 | Table Understanding and (Multimodal) LLMs: A Cross-Domain Case Study on Scientific vs. Non-Scientific Data | TRL Workshop | Introduces a 3,017-table TableEval suite with image, dictionary, HTML, XML, and LaTeX views. | β |
| 2025 | Texts or Images? A Fine-Grained Analysis of Multimodal Table Understanding | Findings of ACL | Breaks down when visual or textual table formats help across table properties and reasoning types. | β |
| 2024 | Tables as Texts or Images: Evaluating Table Reasoning Ability of LLMs and MLLMs | Findings of ACL | Directly compares equivalent tables supplied as serialized text and rendered images. | β |
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2023 | A Study on Reproducibility and Replicability of Table Structure Recognition Methods | ICDAR | Reimplements leading methods under controlled settings and releases GenTSR to test generalization. | β |
| 2023 | Aligning Benchmark Datasets for Table Structure Recognition | ICDAR | Canonicalizes inconsistent annotations across major TSR datasets and measures the effect data alignment has on performance. | π» Code |
Surveys of table recognition, table pre-training, table reasoning, TableQA, and LLM/VLM-based table processing. 7 papers.
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2025 | Large Language Model for Table Processing: A Survey | Frontiers of Computer Science | Surveys LLM- and VLM-based table processing across question answering, image-table extraction, spreadsheet manipulation, training, prompting, and agents. | β |
| 2025 | Table Question Answering in the Era of Large Language Models: A Comprehensive Survey of Tasks, Methods, and Evaluation | arXiv | Reviews TableQA task formulations, benchmarks, reasoning challenges, LLM methods, evaluation practices, and emerging reinforcement-learning directions. | β |
| 2025 | Tabular Data Understanding with LLMs: A Survey of Recent Advances and Challenges | arXiv | Organizes tabular representations and understanding tasks while highlighting complex structures, multimodal inputs, long tables, and multi-table reasoning. | β |
| 2024 | A Review on Table Recognition Based on Deep Learning | arXiv | Reviews end-to-end and modular approaches across table detection, structure, and content recognition. | β |
| 2024 | A Survey of Table Reasoning with Large Language Models | arXiv | Surveys LLM-era table reasoning techniques, analyzes why LLMs improve tabular reasoning, and outlines methodological and application-oriented research directions. | β |
| 2024 | Deep Learning for Table Detection and Structure Recognition: A Survey | ACM Computing Surveys | Surveys datasets, architectures, and evaluation practices for table detection and structure recognition. | π» Code |
| 2022 | Table Pre-training: A Survey on Model Architectures, Pre-training Objectives, and Downstream Tasks | IJCAI Survey Track | Reviews table pre-training architectures, objectives, and downstream tasks including question answering, table search, relation classification, and formula prediction. | β |
Useful adjacent literature is kept directly in the main README for completeness while remaining visually separated from the core taxonomy. 17 papers.
| Year | Paper | Venue | Why it matters | Resources |
|---|---|---|---|---|
| 2025 | OmniParser V2: Structured-Points-of-Thought for Unified Visual Text Parsing and Its Generality to Multimodal Large Language Models | arXiv | Uses structured points-of-thought to unify visual text parsing and transfer document structure skills to MLLMs. | β |
| 2025 | Structure- and Semantic-Based Rationale Distillation: Table and Chart Question Answering in Scientific Documents | RACS | Distills structural and semantic rationales into a compact model for scientific table and chart QA. | β |
| 2025 | Table-R1: Region-based Reinforcement Learning for Table Understanding | arXiv | Injects relevant table regions into reasoning and optimizes region and answer rewards; input tables are serialized rather than visual. | β |
| 2025 | TASER: Table Agents for Schema-guided Extraction and Retrieval | arXiv | Uses agents to extract fragmented document tables into a target schema and retrieve evidence. | β |
| 2025 | Visual Editing as a Chain of Thought for Structured Image Understanding | ICML | REFOCUS iteratively edits structured images to expose relevant evidence, including tables and charts. | β |
| 2025 | WikiMixQA: A Multimodal Benchmark for Question Answering over Tables and Charts | Findings of ACL | Requires multi-hop reasoning over tables and charts distributed across long Wikipedia pages. | β |
| 2024 | Extracting Information from Scientific Literature via Visual Table Question Answering Models | RACS | Compares OCR, visual QA, and structure-preserving pipelines for extracting evidence from scientific tables. | β |
| 2024 | Leopard: A Vision Language Model for Text-Rich Multi-Image Tasks | arXiv | Handles multiple high-resolution text-rich images, with dedicated multi-table and multi-chart evaluation. | β |
| 2024 | mPLUG-DocOwl 1.5: Unified Structure Learning for OCR-free Document Understanding | Findings of EMNLP | A general document VLM with structure-aware pretraining and substantive table recognition and QA evaluation. | β |
| 2024 | Multimodal ArXiv: A Dataset for Improving Scientific Comprehension of Large Vision-Language Models | ACL | Pairs scientific text with figures, tables, and equations for broad document comprehension. | β |
| 2024 | Tabular Insights, Visual Impacts: Transferring Expertise from Tables to Images | ICML | Transfers representations learned from tabular features to ordinary image tasks; related multimodal tabular learning, not visual-table reading. | β |
| 2023 | ICDAR 2023 Competition on Visual Question Answering on Business Document Images | ICDAR | A document-image QA benchmark whose pages frequently require reading tables, forms, and receipts. | β |
| 2023 | Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding | ICML | Pretrains on screenshot-to-HTML parsing and transfers to visually situated language tasks that include tables. | β |
| 2023 | UniMMQA: Unifying Multi-Modal Question Answering with Structured Knowledge | Findings of EMNLP | Unifies QA over text, tables, and ordinary images represented as heterogeneous knowledge sources. | β |
| 2022 | MMCoQA: Conversational Question Answering over Text, Tables, and Images | ACL | Retrieves and answers conversational questions from heterogeneous text, serialized tables, and ordinary images. | β |
| 2022 | OCR-Free Document Understanding Transformer | ECCV | A foundational OCR-free document model used by later visual table recognition and document QA systems. | π» Code |
| 2022 | PP-StructureV2: A Stronger Document Analysis System | arXiv | An end-to-end document analysis system whose SLANet component is widely used for table structure recognition. | π» Code |
- Awesome-Tabular-LLMs β broad table + LLM coverage.
- Awesome-LLM-Table-Mining β table mining with language models.
- Curated Table Structure Recognition β focused TSR bibliography.
- Awesome Table Structure Recognition β TSR papers, datasets, and implementations.