PaddleOCR/latest/version3.x/module_usage/text_recognition #15543
Replies: 17 comments 10 replies
-
你好,我想问一下,PaddleOCR 的示例代码打印的json ,没有检测模型输出的检测框的阈值,类似于detect模型的conf 值,只有rec scores值 |
Beta Was this translation helpful? Give feedback.
-
这上面没有det 检测模型的每个box 框的阈值
At 2025-06-03 13:35:06, "giscus[bot]" ***@***.***> wrote:
PaddleOCR/latest/version3.x/module_usage/text_recognition
Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)
https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/module_usage/text_recognition.html
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
如何自己添加呢,我查看ocr.py 没有看到return 的json 都是继承
在 2025-06-03 14:29:22,"liuhongen1234567" ***@***.***> 写道:
感谢您的反馈,这个确实是当前版本遗漏的一个地方,我们会在下一个wheel版本支持该参数,或者直接使用文本检测模块可以获取该参数,这里dt_scores 就是检测框的阈值。
image.png (view on web)
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
你好,我查看到pipline.py 下保存了rec 的一些信息,是否我可以直接添加det 的信息即可
在 2025-06-03 14:29:22,"liuhongen1234567" ***@***.***> 写道:
感谢您的反馈,这个确实是当前版本遗漏的一个地方,我们会在下一个wheel版本支持该参数,或者直接使用文本检测模块可以获取该参数,这里dt_scores 就是检测框的阈值。
image.png (view on web)
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
直接获取dt_scores 还不是已经排序后的结果吗,我打印了一些,我发现是最后的结果了,
det_scores_list= [item["dt_scores"] foritemindet_results]
在 2025-06-03 15:25:37,"liuhongen1234567" ***@***.***> 写道:
应该不行,检测框还涉及到一个排序的过程,直接添加det信息,识别框和分数可能对应不上
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
我查看下面是直接det 的result 最后的结果了,在往后续都是rec 的存入result 的结果了
在 2025-06-03 15:25:37,"liuhongen1234567" ***@***.***> 写道:
应该不行,检测框还涉及到一个排序的过程,直接添加det信息,识别框和分数可能对应不上
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
这是添加后的值,目前查看应该是没有问题,你这边查看一下,验证是否正确,要和直接检测的示例代码,同步打印一下,但是发现了一个问题,我要去测试det 的conf 是否一直情况,det 示例代码,推理出来的检测效果和我端对端的效果不一致,模型都是同步的
以下是det 示例代码推理结果
在 2025-06-03 15:25:37,"liuhongen1234567" ***@***.***> 写道:
应该不行,检测框还涉及到一个排序的过程,直接添加det信息,识别框和分数可能对应不上
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
主要是用于查看误报框的conf 值为多少,选择合理的thresh 过滤
在 2025-06-03 15:52:41,"liuhongen1234567" ***@***.***> 写道:
获取这个检测分数具体有什么用途吗?如果只是筛选的话可以通过text_det_box_thresh 这个参数筛掉低于阈值的边界框。中间排序过程还比较复杂,不太容易还原出排序后的索引。
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
好的,我想问一下paddlesilm 对paddleocr 的模型进行int8量化,GitHub描述为无损精度量化,是否还可以使用实例代码进行推理预测呢
在 2025-06-03 16:00:15,"liuhongen1234567" ***@***.***> 写道:
您那边是否输入存在一定问题呢?我这边只能看到文字内容。和检测对应不上是正常的,端到端产线和OCR还对边界框进行了一定处理。
image.png (view on web)
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
paddleslim 确保是无损精度下的int8 量化吗,保留精度,速度更快
在 2025-06-03 16:47:58,"liuhongen1234567" ***@***.***> 写道:
应该是可以的,不过paddleslim导出后的模型需要进行一定的修改,这个仓库导出的模型应该不符合现在paddleocr3.0的规范,比如:需要手动添加inference.yaml 等
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
提高paddleocr 的整体推理速度,目前已经使用的trt 和 网络也采用了最轻量级网络,paddlesilm 仓库有点老,估计是不适配v3 模型,和新的推理框架,还有最后一点想问一下您,对于检测模型出现误检测情况,采用box 面积大小过滤,在送入到rec 模型识别,这个在什么位置呢,我看pipline 是没有的
在 2025-06-03 16:56:02,"liuhongen1234567" ***@***.***> 写道:
这个仓库有点老,可能不适用于最新的模型。您那边可以尝试一下。
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
好的,感谢您
在 2025-06-03 17:27:57,"liuhongen1234567" ***@***.***> 写道:
您好, /usr/local/lib/python3.10/dist-packages/paddlex/inference/models/text_detection/processors.py 中的 DBPostProcess 有一个检测框过滤的操作,不知道是不是您所需要的。
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
你好,我想请问一下,有开源rec_mv3_none_none_ctc.yml配置的预训练识别模型吗? |
Beta Was this translation helpful? Give feedback.
-
您好,请问怎么精确获取文本检测、分类、识别的时长呢?谢谢! |
Beta Was this translation helpful? Give feedback.
-
请问PP-OCRv4_server_rec包含哪些层,训练时是否可以冻结指定的层,如何设置? |
Beta Was this translation helpful? Give feedback.
-
您好,训练ocrv4-rec 模型,使用en_dict key 里面不包含任何的符号信息,为什么export 出的inference.yml的dict 却包含了很多符号信息呢。我将符号信息去除模型识别出的结果完全不正确了,这种如何解决呢, 我想保证最后输出的没有任何标点符号信息 |
Beta Was this translation helpful? Give feedback.
-
使用二次开发中的例子微调rec模型会报错,环境: Traceback (most recent call last): |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
PaddleOCR/latest/version3.x/module_usage/text_recognition
Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)
https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/module_usage/text_recognition.html
Beta Was this translation helpful? Give feedback.
All reactions