English

VAREX: A Benchmark for Multi-Modal Structured Extraction from Documents

Computer Vision and Pattern Recognition 2026-04-10 v2

Abstract

We introduce VAREX (VARied-schema EXtraction), a benchmark for evaluating multimodal foundation models on structured data extraction from government forms. VAREX employs a Reverse Annotation pipeline that programmatically fills PDF templates with synthetic values, producing deterministic ground truth validated through three-phase quality assurance. The benchmark comprises 1,777 documents with 1,771 unique schemas across three structural categories, each provided in four input modalities: plain text, layout-preserving text (whitespace-aligned to approximate column positions), document image, or both text and image combined. Unlike existing benchmarks that evaluate from a single input representation, VAREX provides four controlled modalities per document, enabling systematic ablation of how input format affects extraction accuracy -- a capability absent from prior benchmarks. We evaluate 20 models from frontier proprietary models to small open models, with particular attention to models <=4B parameters suitable for cost-sensitive and latency-constrained deployment. Results reveal that (1) below 4B parameters, structured output compliance -- not extraction capability -- is a dominant bottleneck; in particular, schema echo (models producing schema-conforming structure instead of extracted values) depresses scores by 45-65 pp (percentage points) in affected models; (2) extraction-specific fine-tuning at 2B yields +81 pp gains, demonstrating that the instruction-following deficit is addressable without scale; (3) layout-preserving text provides the largest accuracy gain (+3-18 pp), exceeding pixel-level visual cues; and (4) the benchmark most effectively discriminates models in the 60-95% accuracy band. Dataset and evaluation code are publicly available.

Keywords

Cite

@article{arxiv.2603.15118,
  title  = {VAREX: A Benchmark for Multi-Modal Structured Extraction from Documents},
  author = {Udi Barzelay and Ophir Azulai and Inbar Shapira and Idan Friedman and Foad Abo Dahood and Madison Lee and Abraham Daniels},
  journal= {arXiv preprint arXiv:2603.15118},
  year   = {2026}
}

Comments

9 pages, 4 figures, 4 tables, plus 12-page supplementary. Dataset: https://huggingface.co/datasets/ibm-research/VAREX Code: https://github.com/udibarzi/varex-bench

R2 v1 2026-07-01T11:22:03.069Z