-
🔎 Search before asking
🐛 Bug (问题描述)我使用官方给的案例https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/seal_text_detection.html#_3 "dt_scores": [ 🏃♂️ Environment (运行环境)gpu 11.8的dokcer环境,paddle==3.0.0 paddleocr3.0.0 🌰 Minimal Reproducible Example (最小可复现问题的Demo)我使用官方给的案例https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/seal_text_detection.html#_3 图片: |
Beta Was this translation helpful? Give feedback.
Replies: 10 comments 2 replies
-
您好,印章的文本检测主要针对印章类型,对其他文本确实无法兼容,如果是您这张图像,可以用通用的ocr不用印章ocr |
Beta Was this translation helpful? Give feedback.
-
您好,我现在有这样的一个需求,就是想检测一张图片中是否有印章,我用什么方法模型比较好? |
Beta Was this translation helpful? Give feedback.
-
您好,可以直接使用集成了版面区域检测模型、印章检测产线和OCR检测识别模型的PP-StructureV3 https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/pipeline_usage/PP-StructureV3.html#2。 如果只是想要检测图片是否有印章的话可以使用版面区域检测模块 https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/module_usage/layout_detection.html 。 如果检测框中有印章这个类别,则说明存在印章。 |
Beta Was this translation helpful? Give feedback.
-
多谢,我试一下 |
Beta Was this translation helpful? Give feedback.
-
请问哪里有版面检测模型的label含义呢? |
Beta Was this translation helpful? Give feedback.
-
您好,输出结果中有class_id 和 label 信息, ![]() |
Beta Was this translation helpful? Give feedback.
-
不同模型的label_list 不一样,具体可到版面区域文档的模型列表中下载推理模型,然后查看 inference.yaml 中的 label_list 字段。 编号是从0开始的,paragraph_title 对于0, image 对应1, 依次类推 ![]() ![]() |
Beta Was this translation helpful? Give feedback.
-
收到,我现在遇到这种情况,拉取了你们的docker-gpu-11.8的镜像,然后安装了paddleocr-3.0.0,模型启动脚本是 启动时显存占用400M+,在连续调用后显存一直升到3500+且没有释放, |
Beta Was this translation helpful? Give feedback.
-
您好,复现的数据和环境可以提供一下吗?我这边使用demo图片重复100遍,显存一直稳定在600M左右。之前也有用户反馈显存溢出、内存溢出之类的问题,后面他们说是自身环境问题,也没提供复现数据和环境,就不了了之。另外涨到3500 M 也有可能是测试图片非常大,都是超清图片,之后还会接着涨吗? |
Beta Was this translation helpful? Give feedback.
-
I also saw this behaviour and could replicate it. See #15631. I also opened a PR to solve the issue. |
Beta Was this translation helpful? Give feedback.
您好,印章的文本检测主要针对印章类型,对其他文本确实无法兼容,如果是您这张图像,可以用通用的ocr不用印章ocr