PaddleOCR icon indicating copy to clipboard operation
PaddleOCR copied to clipboard

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 de...

Results 1088 PaddleOCR issues
Sort by recently updated
recently updated
newest added

Hi all, I noticed a bug at line 103 of preditc_system.py (paddleocr versoin 2.6) when angle=0 in the below lines: cv_rotate_code = { '90': cv2.ROTATE_90_COUNTERCLOCKWISE, '180': cv2.ROTATE_180, '270': cv2.ROTATE_90_CLOCKWISE }...

- 系统环境/System Environment: - cuda10.2 cudnn8.2 jetpack 4.6.2 tensorrt 8.2.1.8-1 我在jetson nano上通过 `pip3 install paddlepaddle_gpu-2.3.2-cp36-cp36m-linux_aarch64.whl`安装 paddle inference 库,提示安装成功; 然后,通过 `pip3 install "paddleocr>=2.0.1"` 尝试安装paddleocr,但是遇到报错:FileNotFoundError: [Errno 2] No such file or directory:...

依据SVTR的配置训练自己的手写中文数据集(字符个数变动范围:2-15)(输入图像适配:64*512,其他配置不变),训练Loss下降到10左右难以再下降,推理模型发现精度极低,主要为漏字: ![image](https://user-images.githubusercontent.com/12372742/190589637-5d82493c-f74d-44df-b095-06eb071ae04e.png) ![image](https://user-images.githubusercontent.com/12372742/190589692-a3b7f9fd-b341-4787-9e9c-ccb2eba20091.png) ![image](https://user-images.githubusercontent.com/12372742/190589725-b387224f-f56c-4915-ba57-9aa4aba69682.png) 然后,将SVTR中的mixer = Local, 全部换成mixer = Global ,推理发现较为正常,不会出现严重的漏字现象,请问这个现象是否表明Local mixer 导致感受野过大以及丢失的信息较多导致?Local kernel的设置(7, 11)是否也需要结合训练样本的长度来调整?

请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem - 系统环境/System Environment: window10 - 版本号/Version:Paddle: PaddleOCR: 问题相关组件/Related components: - paddleocr 2.5.0.3 paddlepaddle-gpu 2.3.1.post116 - 运行指令/Command Code: python tools/infer/predict_det.py --image_dir="./doc/imgs/0037.jpg"...

| model name | description | model size | download | Update Date | | --------------------------- | ----------------------------------------------------- | ---------- | ------------------------------------------------------------ | ------------- | | ch | Chinese and...

language requests

将CRNN的backbone更换为了VGG,跑通后训练了很久acc一直很低,loss也不怎么下降,这种情况下一步应该考虑是什么问题?主干网络在宽高上是否有其他的限制呢? ![image](https://user-images.githubusercontent.com/58280508/190101066-09536912-2cde-4609-8bb3-f0cb01110e35.png) ![image](https://user-images.githubusercontent.com/58280508/190101248-0e9a06d9-dc06-416a-b9b6-76099dc1e9de.png)

![0018031_1_4_0](https://user-images.githubusercontent.com/20334378/190096349-0dec50eb-30ad-4719-98ba-23330a203f63.jpg) 识别结果就是 ![1663142462778](https://user-images.githubusercontent.com/20334378/190096518-5da54dfc-ac17-4999-845d-81bf5709007f.png) 请问这样的结果应该从哪里去下手分析

你好,我使用相同的数据集,分别用ppocrv2和ppocrv3来finetune,都加载了各自的预训练模型,可以看到在最开始的step里,ppocrv2 acc 是0.29,而ppocrv3 acc是0。下面的两张图分别是v2和v3的。 ![image](https://user-images.githubusercontent.com/14170291/195490471-b7164397-110a-4554-999c-042bb8d73ba4.png) ![image](https://user-images.githubusercontent.com/14170291/195490492-cd3e4676-3b1a-4db2-9bc1-991c2539de75.png) 但是讲道理v3加载预训练模型之后,在最初的step里acc应该超过v2的才对。所以怀疑是不是v3开源的预训练模型有问题,麻烦check一下,谢谢!

请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem - 系统环境/System Environment: - 版本号/Version:Paddle: PaddleOCR: 问题相关组件/Related components: - 运行指令/Command Code: - 完整报错/Complete Error Message: