How to Speed Up Recognition Inference Time?

[tyasnk]

Hi,
I use the following code

from paddleocr import PaddleOCR
model_ocr = PaddleOCR(lang='en', use_gpu=False)
model_ocr.ocr("letter.jpg")

and got more than 1 second on recognition phase

[2022/07/14 21:05:40] ppocr WARNING: Since the angle classifier is not initialized, the angle classifier will not be used during the forward process
[2022/07/14 21:05:41] ppocr DEBUG: dt_boxes num : 11, elapse : 0.35867905616760254
[2022/07/14 21:05:42] ppocr DEBUG: rec_res num  : 11, elapse : 1.218210220336914
[[[[37.0, 55.0], [308.0, 58.0], [307.0, 97.0], [36.0, 94.0]],
  ('Post the letter', 0.973766028881073)],
 [[[180.0, 179.0], [311.0, 179.0], [311.0, 201.0], [180.0, 201.0]],
  ('Sam Brown', 0.9435603618621826)],
...

is there a trick to speed up the recognition time without a GPU (use CPU only)? I want it to below 500ms
I run it on local PC with AMD Ryzen 5 3600 processor

Thanks

[tink2123]

You can try to use intel MKL to speed up:

from paddleocr import PaddleOCR
model_ocr = PaddleOCR(lang='en', use_gpu=False, enable_mkldnn=True)
model_ocr.ocr("letter.jpg")

If an error is reported you can google how to configure MKL for AMD CPU.

[tyasnk]

thanks!