The MERIT Dataset 🎒📃🏆

The MERIT Dataset is a multimodal dataset (image + text + layout) designed for training and benchmarking Large Language Models (LLMs) on Visually Rich Document Understanding (VrDU) tasks. It is a fully labeled synthetic dataset generated using our opensource pipeline available on GitHub. You can explore more details about the dataset and pipeline reading our paper.

Introduction ℹ️

AI faces some dynamic and technical issues that push end-users to create and gather their own data. In addition, multimodal LLMs are gaining more and more attention, but datasets to train them might be improved to be more complex, more flexible, and easier to gather/generate.

In this research project, we identify school transcripts of records as a suitable niche to generate a synthetic challenging multimodal dataset (image + text + layout) for Token Classification or Sequence Generation.

Hardware ⚙️

We ran the dataset generator on an MSI Meg Infinite X 10SF-666EU with an Intel Core i9-10900KF and an Nvidia RTX 2080 GPU, running on Ubuntu 20.04. Energy values in the table refer to 1k samples, and time values refer to one sample.

Task	Energy (kWh)	Time (s)
Generate digital samples	0.016	2
Modify samples in Blender	0.366	34

Benchmark 💪

We train the LayoutLM family models on Token Classification to demonstrate the suitability of our dataset. The MERIT Dataset poses a challenging scenario with more than 400 labels.

We benchmark on three scenarios with an increasing presence of Blender-modified samples.

• Scenario 1: We train and test on digital samples.
• Scenario 2: We train with digital samples and test with Blender-modified samples.
• Scenario 3: We train and test with Blender-modified samples.

	Scenario 1	Scenario 2	Scenario 3	FUNSD/	Lang.	(Tr./Val./Test)
	Dig./Dig.	Dig./Mod.	Mod./Mod	XFUND
	F1	F1	F1	F1
LayoutLMv2	0.5536	0.3764	0.4984	0.8276	Eng.	7324 / 1831 / 4349
LayoutLMv3	0.3452	0.2681	0.6370	0.9029	Eng.	7324 / 1831 / 4349
LayoutXLM	0.5977	0.3295	0.4489	0.7550	Spa.	8115 / 2028 / 4426

Citation

If you find our research interesting, please cite our work. 📄✏️

@article{de2024merit,
  title={The MERIT Dataset: Modelling and Efficiently Rendering Interpretable Transcripts},
  author={de Rodrigo, I and Sanchez-Cuadrado, A and Boal, J and Lopez-Lopez, AJ},
  journal={arXiv preprint arXiv:2409.00447},
  year={2024}
}