| TinyOCR |
| Copyright 2026 Maureese Williams |
|
|
| This product includes PaddleOCR / PP-OCRv6 model artifacts. |
| Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved. |
| https://github.com/PaddlePaddle/PaddleOCR |
| Licensed under the Apache License, Version 2.0 (see LICENSE). |
|
|
| PP-OCRv6_det_tiny.onnx is the unchanged pretrained detector. |
| PP-OCRv6_rec_tiny.onnx WAS MODIFIED by Maureese Williams in 2026: |
| selected float32 weight tensors were stored as per-channel int8 values with |
| constant decoding back to float32. Computation remains float32. No training |
| or calibration was performed. This modified file is not an official |
| PaddlePaddle release. SHA-256 hashes and origins are in assets/models.json. |
|
|
| Original ONNX artifacts were distributed through RapidAI/RapidOCR v3.10.0: |
| https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv6/det/PP-OCRv6_det_tiny.onnx |
| https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv6/rec/PP-OCRv6_rec_tiny.onnx |
| Official Apache-2.0 model cards: |
| https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_det_onnx |
| https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_rec_onnx |
|
|
| Runtime inference uses the separately installed RapidOCR project (Apache-2.0): |
| Copyright RapidAI contributors. https://github.com/RapidAI/RapidOCR |
| Other installed dependencies retain their own licenses. |
|
|
| Research repository only: the vendored Google HierText evaluator has its own |
| CC BY-SA 4.0 license and notices in third_party/hiertext. It and evaluation |
| datasets are not included in the TinyOCR runtime distribution. |
|
|
