PaddleOCR/latest/en/index

[giscus[bot]]

PaddleOCR/latest/en/index

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/en/index.html

[roostaeecs]

I am using PaddleOCR with the Arabic language model (lang='fa') to perform OCR on Arabic images. While PaddleOCR correctly recognizes the Arabic characters, it processes the text in a Left-to-Right (LTR) order, which is incorrect for Arabic, a Right-to-Left (RTL) language. This results in the words and sentences being in reverse order.

I have reviewed the paddleocr --help output to see if there are any options to explicitly set the text direction or handle RTL languages like Arabic.

My question is:

Is there a specific option in PaddleOCR, possibly using ocr_order_method or another parameter, to correctly handle Right-to-Left languages like Arabic and ensure the output text is in the correct RTL order?

If there isn't a built-in option, what are the recommended workarounds to post-process the OCR output to reorder the text correctly for RTL languages in Python?

Any guidance or solutions on how to get PaddleOCR to output Arabic text in the correct Right-to-Left order would be greatly appreciated.

I tried to use the following code:

from paddleocr import PaddleOCR, draw_ocr

ocr = PaddleOCR(use_angle_cls=True, lang='Ar')
img_path = 'image5.jpg'
result = ocr.ocr(img_path, cls=True)
for idx in range(len(result)):
res = result[idx]
for line in res:
print(line)

draw result

from PIL import Image
result = result[0]
image = Image.open(img_path).convert('RGB')
boxes = [line[0] for line in result]
txts = [line[1][0] for line in result]
scores = [line[1][1] for line in result]
im_show = draw_ocr(image, boxes, txts, scores, font_path='./doc/fonts/arabic.ttf')
im_show = Image.fromarray(im_show)
im_show.save('result.jpg')

[PengqiH]

.Warning: tiling offset out of range, index: 32
你好，这个是什么问题啊，如果是第一次加载，或者是更换一张没识别过的图片，就会显示这个错误，占用非常久的时间

[liuhongen1234567]

您好，这个可以提供具体的复现命令吗？

[PengqiH]

你好，指令是：paddlex --serve --pipeline table_recognition_v2 --port 8082 --device npu:0，使用表格分类的，在华为昇腾910B显卡上运行，但是发现一个问题：①有时候会显示.Warning: tiling offset out of range, index: 32 ②切换图片识别的时候(如果之前没有识别过)，则速度会非常慢，少则10秒，多则几百秒 ③怀疑是不是和warning的信息有关，如果有关，方便说明一下该如何处理嘛

[liuhongen1234567]

您好，可以参考一下这个issue https://github.com/PaddlePaddle/Paddle/issues/68107， export FLAGS_npu_jit_compile=0 试试呢？看起来是npu的一个bug

[PengqiH]

你好，这个已经试过了，但还是会出现对应的warning，而且运行速度比较慢，pipeline中是有十个模型，但是最短的用时是三秒左右，长的则是五六秒，这个是正常的嘛

[liuhongen1234567]

您好，推理速度和pipeline中的模型和输入图片大小有关，如果模型比较多的话。5-6s也算正常吧，可以测一下GPU的速度。 这个速度慢似乎和npu算子的编译机制有关，这个我也没有好的办法。

[hzdavid]

great

[PaoloP-creator]

/Users/home/Downloads/汤用彤 - 汉魏两晋南北朝佛教史 2.pdf

[pamlelr98]

Hello, I have already trained the model using my own dataset and custom dictionary file, specifically vi_dict in PP-OCRv5_server_rec. However, when I try to run prediction, I couldn’t find any way to specify the path to my custom dictionary file, like we could with the rec_char_dict_path parameter in OCRv4.

It seems that PP-OCRv5 no longer supports this parameter, or at least I couldn’t find any documentation about it. How can I load and use my custom dictionary file during prediction in PP-OCRv5?

Thank you very much!

[liuhongen1234567]

Hello, the dictionary for PP-OCRv5 is stored in the inference.yaml file of the exported model. During model inference, it is read directly and does not need to be manually specified.

[samjia533332]

hi I am running PPStructureV3 on my mac local
but I got errors:

All the weights of PPChart2TableInference were initialized from the model checkpoint at /Users/SamJia/.paddlex/official_models/PP-Chart2Table.
If your task is similar to the task the model of the checkpoint was trained on, you can already use PPChart2TableInference for predictions without further training.
Loading configuration file /Users/SamJia/.paddlex/official_models/PP-Chart2Table/generation_config.json
Segmentation fault: 11
(base) 192:experiments-table SamJia$ /Users/SamJia/anaconda3/lib/python3.12/multiprocessing/resource_tracker.py:254: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown
warnings.warn('resource_tracker: There appear to be %d '

Can anyone help?

[liuhongen1234567]

Hello, this issue is usually caused by PPChart2Table having too high memory usage, leading to a memory crash. You can try setting use_chart_recognition=False when initializing pipeline = PPStructureV3() to disable that pipeline.