BaFCo

BaFCo: A Document Understanding Benchmark for Complex Bangla Form Comprehension (ECCV 2026) Real Bangladeshi government forms with manual annotations for two tasks: Document Layout Analysis (DLA) and Key Information Extraction (KIE). This is the dataset and its schema. The full benchmark pipeline (inference, evaluation, and post-processing) lives in the code repository: github.com/mausulazad/BaFCo. Paper: arXiv:2607.05614. BaFCo curates 200 complex multi-page government forms… See the full description on the dataset page:

Source: Hugging Face Hub (Mausul/bafco). Metadata imported from the dataset’s Hub tags.