changdazhou comments

Results 132 comments of


                                            changdazhou

When there is tps in the network, variable length input is not supported, and the input size needs to be the same as during training

Thank you for your feedback. According to the log information, the current version does indeed not support processing variable-length input when there is TPS in the network. It is recommended...

PaddleOCR-VL后处理合框逻辑问题

1. 并不少，这里的判断是针对插图采用"文字环绕"的情形时，导致段落被分成了两个block的情况，这里判断重叠的框就是指那个插图。 2. 这个作用就是对于双栏的左下框和右上检测框进行合并，然后采用居中对齐的方式将两个子图拼接，让VL模型来决定这两段文本是否是真的一个段落，由于VL模型不会返回内容坐标，所以从结果上来看无论这两个框的内容是否合并，json中都会将后一个框的内容放到前一个框里面。

fix: convert numpy array to list to avoid AttributeError on append

感谢反馈，该问题已在#4787中得到修复，这个问题的本质是上游单模块模型推理结果数据结构不确定导致的，不是此处的数据类型不对。

ModuleNotFoundError: No module named 'langchain.docstore'

Thank you for your submission. Please ensure your code is submitted to the `develop` branch, and after the code is merged, we will synchronize it to the `release` branch. Additionally,...

产品线 table_recognition，官方示例中代码中错误。 ocr_res 变量没有初始化。

请问是使用的最新代码吗，建议更新到最新代码测试一下哈，该BUG已修复

Thank you for your feedback. For PaddleOCR versions 3.0 and above, we recommend referring to the new documentation: https://github.com/PaddlePaddle/PaddleOCR/blob/main/docs/version3.x/deployment/serving.en.md. Please confirm your version and follow the new documentation.

changdazhou

使用PPStructureV3进行公式识别报错

使用PPStructureV3进行公式识别报错

When there is tps in the network, variable length input is not supported, and the input size needs to be the same as during training

PaddleOCR-VL服务化部署解析时存在空的情况

PaddleOCR-VL后处理合框逻辑问题

是否支持在ARM架构下编译部署

fix: convert numpy array to list to avoid AttributeError on append

ModuleNotFoundError: No module named 'langchain.docstore'

产品线 table_recognition，官方示例中代码中错误。 ocr_res 变量没有初始化。

docker build fail