This repository is a curated list of deep learning-based papers for text detection and recognition in natural scenes, serving researchers and practitioners in the OCR field. It provides a structured overview of state-of-the-art methods, their performance metrics, and associated code or trained models, facilitating rapid evaluation and adoption of relevant techniques.

How It Works

The list categorizes papers by task (detection, recognition, end-to-end) and sorts them by publication date. Each entry includes conference/journal, date, title, performance scores (e.g., F1, word accuracy) on benchmark datasets like ICDAR (IC), and resource availability (official code, trained models). This structured approach allows users to quickly compare different methodologies and their reported effectiveness.

Quick Start & Requirements

This is a curated list, not a runnable software package. To use specific methods, users must refer to the individual paper's repository or implementation details linked within the list. Requirements vary per paper, often including Python, deep learning frameworks (TensorFlow, PyTorch, Caffe, MXNet), and specific hardware (GPUs).

Highlighted Details

• Comprehensive coverage of text detection and recognition papers from major conferences (CVPR, ICCV, ECCV, AAAI, etc.) and journals.
• Performance scores (F1, word accuracy) are provided for key datasets (ICDAR, SVT, IIIT5k), enabling direct comparison of methods.
• Indicates availability of official code (*CODE) and pre-trained models (CODE(M)) for many entries, facilitating practical implementation.
• Includes links to related resources such as project pages, demos, slides, and tutorials.

Maintenance & Community

This repository is maintained by the OCR team at Clova AI, powered by NAVER-LINE. It is scheduled for regular updates following major AI conferences. Community resources like Discord/Slack are not explicitly mentioned.

Licensing & Compatibility

The repository itself is a list and does not have a specific license. Individual papers and their associated code/models will have their own licenses, which users must consult. Compatibility for commercial use depends entirely on the licenses of the linked resources.

Limitations & Caveats

The list primarily focuses on papers published up to early 2020, with limited coverage of more recent advancements. Performance scores are reported as published and may not reflect current state-of-the-art or standardized evaluation protocols. The "Others Papers" section is less structured, covering diverse sub-tasks like dataset creation and document analysis.