How to use Recognition Only in Paddle OCR

[ahmadmustafaanis]

Hi, I do not want my OCR to do detection, I want it to do recognition only, but it only returns a single character.

How can I only run recognizer only, and not detector.

[matteoPiccoli]

@ahmadmustafaanis

First initialize your model with:

Paddle = paddleocr.PaddleOCR(use_angle_cls=True, lang=your_language, use_gpu=if_gpu_available)

Then call the recognition only by setting the flags as follows:

result = Paddle.ocr(image, cls=False, det=False, rec=True)

In the result variable you can find all your text OCR predictions.

[ahmadmustafaanis]

@matteoPiccoli

Here are my results when I am using Det + Rec

Here are my results when I am using Rec only on the same image.

My ocr initialization is

ocr = PaddleOCR(
    use_angle_cls=False,
    lang='en', 
    table=False, 
    use_mp=True,
    # image_dir='ocr_images',
    enable_mkldnn=True,
    use_gpu=True,
    max_batch_size = 20,
    total_process_num = os.cpu_count() * 2 - 1,
    )

[matteoPiccoli]

@ahmadmustafaanis

Right now I'm doing a similar thing in my project. Try to detect the text first and then apply the recognition only on the cropped regions of the image contained in the predicted bounding boxes.

Adapt the following pseudo-code according to your needs:

import cv2
from paddleocr import PaddleOCR

Paddle = PaddleOCR(use_angle_cls=True, lang="your_language", use_gpu=True)
result = Paddle.ocr(img_path, cls=False, det=True, rec=False)

img= cv2.imread(img_path)
boxes = []
for line in result:
        boxes.append([[int(line[0][0]), int(line[0][1])], [int(line[2][0]), int(line[2][1])]])

for box in boxes:
        cropped_image = copy[box[0][1]:box[1][1], box[0][0]:box[1][0]]
        rec_result = Paddle.ocr(cropped_image, cls=False, det=False, rec=True)
        text, _ = rec_result[0]

[rahmatdinar]

can we just disable the ppocr DEBUG printout on terminal?

[Lahari-77]

yes just set show_log parameter to False during initialisation of paddleocr.