How can I make it return every word instead of word groups?

[Maxhem2]

I'm working with the example but can't figure out how to get single words instead of word groups.

from paddleocr import PaddleOCR, draw_ocr

ocr = PaddleOCR(use_angle_cls=True, lang='en')
img_path = r'C:\Users\UserA\image.jpg'
result = ocr.ocr(img_path, cls=True)
for idx in range(len(result)):
    res = result[idx]
    for line in res:
        print(line)

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='C:/Users/UserA/PaddleOCR-2.10.0/doc/fonts/simfang.ttf')
im_show = Image.fromarray(im_show)
im_show.save('result.jpg')

[GreatV]

To make PaddleOCR return single words instead of word groups, you can modify the way you process the OCR output. By default, PaddleOCR detects and returns text groups consisting of multiple words in a single bounding box. To extract single words, you need to split detected text into individual words manually.

Solution:

Modify your code so that it processes each detected text block and splits them into individual words.

Updated Code:

from paddleocr import PaddleOCR, draw_ocr
from PIL import Image

# Initialize OCR
ocr = PaddleOCR(use_angle_cls=True, lang='en')

# Image path
img_path = r'C:\Users\UserA\image.jpg'

# Run OCR
result = ocr.ocr(img_path, cls=True)

# Process results
for idx in range(len(result)):
    res = result[idx]
    for line in res:
        bbox, (text, score) = line  # Bounding box and recognized text
        words = text.split()  # Split text into individual words
        for word in words:
            print(f"Word: {word}, Confidence: {score}")

# Visualization
image = Image.open(img_path).convert('RGB')
boxes = [line[0] for line in result[0]]  # Extract bounding boxes
txts = [word for line in result[0] for word in line[1][0].split()]  # Extract individual words
scores = [line[1][1] for line in result[0]]  # Extract confidence scores

# Draw OCR result
im_show = draw_ocr(image, boxes, txts, scores, font_path='C:/Users/UserA/PaddleOCR-2.10.0/doc/fonts/simfang.ttf')
im_show = Image.fromarray(im_show)
im_show.save('result.jpg')

Explanation:

1. Splitting Detected Phrases: The recognized text (line[1][0]) is split into individual words using split().
2. Processing Individual Words: Instead of printing a phrase, the loop iterates over individual words.
3. Updating txts for Visualization: The txts list is updated to contain single words rather than full phrases.

Alternative Approach:

If you need precise word-level bounding boxes, you might need to modify the OCR model or explore a character-level or word-level detection approach. A related discussion on how to get single characters and coordinates is available here:
👉 GitHub Discussion - PaddleOCR: How to Get Single Character and Its Coordinates

Would you like more details on fine-tuning the model to enhance word-level detection? 😊

---

Response generated by 🤖 feifei-bot | chatgpt-4o-latest

⚠️ This is an AI-generated response. Please verify any critical information.

[Maxhem2]

Thanks AI response, but unfortunately I get this error.

Traceback (most recent call last):
  File "c:\Users\UserA\test2.py", line 29, in <module>
    im_show = draw_ocr(image, boxes, txts, scores, font_path='C:/Users/UserA/PaddleOCR-2.10.0/doc/fonts/simfang.ttf')
  File "c:\Users\UserA\AppData\Local\Programs\Python\Python310\lib\site-packages\paddleocr\tools\infer\utility.py", line 481, in draw_ocr
    txt_img = text_visual(
  File "c:\Users\UserA\AppData\Local\Programs\Python\Python310\lib\site-packages\paddleocr\tools\infer\utility.py", line 619, in text_visual
    assert len(texts) == len(
AssertionError: The number of txts and corresponding scores must match

[GreatV]

please refer to https://paddlepaddle.github.io/PaddleOCR/latest/en/ppstructure/blog/return_word_pos.html