OpenMMLab Text Detection, Recognition and Understanding Toolbox

A curated list of resources for Document Understanding (DU) topic

Code for the paper "PICK: Processing Key Information Extraction from Documents using Improved Graph Learning-Convolutional Networks" (ICPR 2020)

Algorithms, papers, datasets, performance comparisons for Document AI.

Doc2Graph transforms documents into graphs and exploit a GNN to solve several tasks.

Key information extraction from invoice document with Graph Convolution Network

An unofficial PyTorch implementation of "Lin et al. ViBERTgrid: A Jointly Trained Multi-Modal 2D Document Representation for Key Information Extraction from Documents. ICDAR, 2021"

[MM'2024] PEneo, an effective algorithm for key-value pair extraction from form-like documents, designed for real-world applications.

The task aims at extracting required fields in receipts captured by mobile devices 😄

[MM'2024] Official release of RFUND introduced in the MM'2024 paper "PEneo: Unifying Line Extraction, Line Grouping, and Entity Linking for End-to-end Document Pair Extraction"

利用llm大语言模型提取卡证票据关键信息。Key Information Extraction from Image with LLM(large language model).Basically, it can extract key information from all bills and documents.

Public benchmark for document data extraction - hand-verified labels, schemas, a standalone scorer, and per-document results across 50 hard public documents.

Self-hosted AI accountant. Reads receipts / invoices, reconciles ledgers, checks its own numbers, and asks before anything irreversible. Open source, agentic, runs on your infrastructure.

A new package would process user-provided text inputs, such as headlines or short descriptions, and generate structured summaries or categorizations using an LLM. It would be particularly useful for c

news-summizr extracts structured summaries from headlines, labeling key points like announcement, products, region for quick insight.

Benchmarked invoice and document extraction pipeline: PII redaction before any model call, layout-aware extraction into a strict schema, field-level confidence scoring, validation rules, a human review queue, and corrections fed back as exemplars. Runs offline, no API key.

A new package that processes news headlines or short text inputs to generate structured summaries of events, such as service disruptions or incidents. It uses an LLM to extract key details like the co

A new package designed to facilitate the extraction of structured summaries or key information from user inputs related to Unix Fourth Edition. It processes textual prompts about Unix concepts, comman

AI & ML research project for automatic product extraction, classification, and analysis of receipt data

Benchmarking agentic and single-pass extraction strategies across LLM providers on the Kleister NDA dataset