ocr result appear other language word

[zhujun5164]

Hi

i have set the langs to 'zh' when i using the OCR，but the recognize result appear japanese word or other language word。how can i fix it or limit the ocr result in my word dict.

thx

[zhujun5164]

code

from PIL import Image
from surya.ocr import run_ocr
from surya.model.detection.model import load_model as load_det_model, load_processor as load_det_processor
from surya.model.recognition.model import load_model as load_rec_model
from surya.model.recognition.processor import load_processor as load_rec_processor
from surya.recognition import batch_recognition

image = Image.open('text.png')
langs = ["en"] # Replace with your languages - optional but recommended
det_processor, det_model = load_det_processor(), load_det_model()
rec_model, rec_processor = load_rec_model(), load_rec_processor()

predictions = batch_recognition([image], [langs], rec_model, rec_processor)
print(predictions)

image

output
([' ସଙ୍କିତ '], [0.77392578125])

[EHadoux]

Well, I actually came here to say the same but with english. I deal with financial reports and it often changes the £ with "છ" or "רא" or "મ" or "દ".
I must say in all fairness that the rest is great.

[EHadoux]

If that helps, here is an example with docs where 594 is read as રેત્વે but the rest is near perfect.

input.pdf
output.md

It's using marker but I feel like it's a surya issue. (v0.8.3)

!PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 TORCH_DEVICE=mps OCR_ENGINE=surya marker_single --languages en --paginate_output --force_ocr --output_format markdown ./input.pdf --output_dir ./MarkerOutput/

EDIT: I've tested the latest version of surya by itself (0.9.1), and it looks like the problem is not there anymore