mathieu1256/FATURA2-invoices

The dataset consists of 10000 jpg images with white backgrounds, 10000 jpg images with colored backgrounds (the same colors used in the paper) as well as 3x10000 json annotation files. The images are generated from 50 different templates.

https://zenodo.org/records/10371464

---

dataset_info: features: - name: image dtype: image - name: ner_tags sequence: int64 - name: words sequence: string - name: bboxes sequence: sequence: int64 splits: - name: train num_bytes: 477503369.0 num_examples: 10000 download_size: 342662174 dataset_size: 477503369.0 configs: - config_name: default data_files: - split: train path: data/train-*

@misc{limam2023fatura, title={FATURA: A Multi-Layout Invoice Image Dataset for Document Analysis and Understanding}, author={Mahmoud Limam and Marwa Dhiaf and Yousri Kessentini}, year={2023}, eprint={2311.11856}, archivePrefix={arXiv}, primaryClass={cs.CV} }