Releases: breezedeus/CnSTD
Releases · breezedeus/CnSTD
Bug Fixed
Integrated the latest PPOCRv5 text detection models
Update 2025.06.25:发布 V1.2.6
Major Changes:
- Integrated the latest PPOCRv5 text detection functionality based on RapidOCR for even faster inference speed
- Added support for PP-OCRv5 detection models:
ch_PP-OCRv5_detandch_PP-OCRv5_det_server
- Added support for PP-OCRv5 detection models:
- Fixed some known bugs
主要变更:
- 基于 RapidOCR 集成 PPOCRv5 最新版文本检测功能,提供更快的推理速度
- 新增支持 PP-OCRv5 检测模型:
ch_PP-OCRv5_det和ch_PP-OCRv5_det_server
- 新增支持 PP-OCRv5 检测模型:
- 修复部分已知 bug
Bug Fixes
Update 2024.12.08:发布 V1.2.5.2
Bug Fixes:
- Fix compatibility issue of setting environment variables on Windows systems
- Use subprocess.run instead of os.system for better cross-platform support
Bug Fixes:
- 修复在 Windows 系统下设置环境变量的兼容性问题
- 使用 subprocess.run 替代 os.system 以提供更好的跨平台支持
Update 2024.11.30:发布 V1.2.5.1
Update 2024.11.30:发布 V1.2.5.1
Major Changes:
- en_PP-OCRv3_det still uses the previous version and does not use RapidDetector
Bug Fixes:
- en_PP-OCRv3_det 依旧使用之前的版本,不使用 RapidDetector
v1.2.5: Merge pull request #88 from breezedeus/pytorch
Update 2024.11.24:发布 V1.2.5
Major Changes:
- Integrated latest PPOCRv4 text detection functionality based on RapidOCR for faster inference
- Added support for PP-OCRv4 detection models, including standard and server versions
- Added support for PP-OCRv3 English detection model
- Optimized model download functionality with support for domestic mirrors
主要变更:
- 基于 RapidOCR 集成 PPOCRv4 最新版文本检测功能,提供更快的推理速度
- 新增支持 PP-OCRv4 检测模型,包括标准版和服务器版
- 新增支持 PP-OCRv3 英文检测模型
- 优化模型下载功能,支持从国内镜像下载模型文件
新模型名称:
model_name |
PyTorch 版本 | ONNX 版本 | 支持检测的语言 | 模型文件大小 |
|---|---|---|---|---|
| en_PP-OCRv3_det | X | √ | 英文、数字 | 2.3 M |
| ch_PP-OCRv4_det | X | √ | 简体中问、英文、数字 | 4.5 M |
| ch_PP-OCRv4_det_server | X | √ | 简体中问、英文、数字 | 108 M |
使用方式
- 命令行:
$ cnstd predict -m ch_PP-OCRv4_det_server -i examples/1_res.png -o outputs - Python:
from cnstd import CnStd std = CnStd(model_name="ch_PP-OCRv4_det_server") out = std.detect("examples/1_res.png") print(out)
Fix: some formats of models require fixed-size input images
Update 2024.06.22:发布 V1.2.4.2
Major Changes:
- Added a new parameter
static_resized_shapewhen initializingYoloDetector, which is used to resize the input image to a fixed size. Some formats of models require fixed-size input images during inference, such asCoreML.
主要变更:
YoloDetector初始化时加入了参数static_resized_shape, 用于把输入图片 resize 为固定大小。某些格式的模型在推理时需要固定大小的输入图片,如CoreML。
fix: a bug in the detect method of YoloDetector
Update 2024.06.17:发布 V1.2.4.1
Major Changes:
- Fixed a bug in the
detectmethod ofYoloDetector: when the input is a single file, the output is not a double-layer nested list.
主要变更:
- 修复了
YoloDetector中detect方法的一个bug:输入为单个文件时,输出不是双层嵌套的 list。
Support for YOLO Detector based on Ultralytics.
Update 2024.06.16:发布 V1.2.4
Major Changes:
- Support for YOLO Detector based on Ultralytics.
主要变更:
- 支持基于 Ultralytics 的 YOLO Detector。
V1.2.3.6: 修改默认的模型下载方式
Update 2024.04.10:发布 V1.2.3.6
主要变更:
- CN OSS 不可用了,默认下载模型地址由
CN改为HF。
use environment variable to determine the download source for models
Main Changes:
-
Added support for the
CNSTD_DOWNLOAD_SOURCEenvironment variable to determine the download source for models. The default source is the Aliyun OSS address. -
In
LayoutAnalyzer, introduced two new parameters,model_categoriesandmodel_arch_yaml, to specify the list of model categories and model architectures.