easyocr_deploy
Copyright 2020-2024 Jaided AI
Copyright 2019-present NAVER Corp.

This product includes software developed by Jaided AI and published as EasyOCR
(https://github.com/JaidedAI/EasyOCR), licensed under the Apache License,
Version 2.0; see the LICENSE file in this directory.

The published weights are release assets of that same Apache-2.0 repository.

Two upstream projects meet inside EasyOCR, and both are acknowledged here
because the code in this directory is derived from each of them.

  - CRAFT: Character Region Awareness for Text Detection
    Copyright (c) 2019-present NAVER Corp.
    https://github.com/clovaai/CRAFT-pytorch
    MIT License

    The detector's architecture and its heatmap postprocessing come from here.
    Baek, Lee, Han, Yun, Lee, "Character Region Awareness for Text Detection",
    CVPR 2019 (arXiv:1904.01941).

    NAVER published the trained weights (craft_mlt_25k.pth) from Google Drive
    without stating terms for them separately from the MIT-licensed code. The
    file redistributed with mozo is the copy Jaided AI publishes as a release
    asset of its Apache-2.0 repository, and that is the chain relied on. This is
    recorded rather than glossed: it is one link weaker than the other families
    here, where the publisher states the checkpoint's licence directly.

  - deep-text-recognition-benchmark
    Copyright (c) 2019-present NAVER Corp.
    https://github.com/clovaai/deep-text-recognition-benchmark
    Apache License, Version 2.0

    The recogniser's architecture -- a convolutional stack, two bidirectional
    LSTMs and a CTC head -- comes from here. Baek et al., "What Is Wrong With
    Scene Text Recognition Model Comparisons?", ICCV 2019 (arXiv:1904.01906).
    The published weights were retrained by Jaided AI, per script.

NOT INCLUDED
------------
The training code, the DBNet detector, the beam-search and word-beam-search
decoders, paragraph grouping, right-to-left display reordering, batched
multi-image inference, arbitrary-rotation retries and the first-generation
(ResNet) recogniser are all left behind. See PROVENANCE.md for the full list
and for why each one is out.
