Paddle OCR not working for live video feed

[Adhitya1996]

Issue: I have experimented a lot for several weeks but I am pasting a simple code version to understand if logic is failing at any point. I request the community's help at the earliest. Ofc, Paddle OCR worked well with images and I was able to fetch very good results but I need it to work for live video feed.

Error:
Traceback (most recent call last):
File "D:\01_IAV-Codes\02_OCR-Models\test_01.py", line 22, in
frame = draw_ocr(frame, line, font_path=font_path)
File "C:\Users\I011786\Miniconda\envs\env-paddle\lib\site-packages\paddleocr\tools\infer\utility.py", line 384, in draw_ocr
scores = [1] * len(boxes)
TypeError: object of type 'NoneType' has no len()

My code:
import cv2
from paddleocr import PaddleOCR, draw_ocr
from PIL import Image, ImageFont

Start video capture

cap = cv2.VideoCapture(0)

Use a default font

font_path = ImageFont.load_default()
ocr = PaddleOCR(use_gpu=True, use_angle_cls=True, lang='en') # need to run only once to download and load model into memory
while True:
# Capture frame-by-frame
ret, frame = cap.read()
if not ret:
break

# Perform OCR on the frame
result = ocr.ocr(frame)

# Display the resulting frame with bounding boxes and detected texts
if result:
    for line in result:
        frame = draw_ocr(frame, line, font_path=font_path)

# Only for images
# for idx in range(len(result)):
#     res = result[idx]
#     print("\n Line: \n")
#     for line in res:
#         print(line)

# Display the frame
cv2.imshow("Video", frame)

if cv2.waitKey(1) & 0xFF == ord('q'):
    break

cap.release()
cv2.destroyAllWindows()

[Sunting78]

Hello, please check if the result is normal. You can print and paste the result here for a look.

[Adhitya1996]

Hey, thanks a lot for replying but the result is not printed at all. Issue I guess is with video frames. It worked well for images for me because when images are attached as JPEG or PDF, there are no empty values. But for video frames, it is probably giving out empty arrays and I tried to solve around it but it didn't work for me so far. Hence, this issue: TypeError: object of type 'NoneType' has no len().

Code:

import cv2
from paddleocr import PaddleOCR, draw_ocr
from PIL import Image, ImageFont

Start video capture

cap = cv2.VideoCapture(0)

Use a default font

font_path = ImageFont.load_default()
ocr = PaddleOCR(use_gpu=True, use_angle_cls=True, lang='en') # need to run only once to download and load model into memory
while True:
# Capture frame-by-frame
ret, frame = cap.read()
if not ret:
break

# Perform OCR on the frame
result = ocr.ocr(frame)
print("Result 1")
print(result)

# Display the resulting frame with bounding boxes and detected texts
if result:
    for line in result:
        frame = draw_ocr(frame, line, font_path=font_path)
        print("Result 2")
        print(result)

# Only for images
# for idx in range(len(result)):
#     res = result[idx]
#     print("\n Line: \n")
#     for line in res:
#         print(line)

# Display the frame
cv2.imshow("Video", frame)

if cv2.waitKey(1) & 0xFF == ord('q'):
    break

cap.release()
cv2.destroyAllWindows()

Output:

[2024/04/18 10:25:20] ppocr DEBUG: dt_boxes num : 0, elapsed : 0.9545092582702637
[2024/04/18 10:25:20] ppocr DEBUG: cls num : 0, elapsed : 0
[2024/04/18 10:25:20] ppocr DEBUG: rec_res num : 0, elapsed : 0.0
Result 1
[None]
Traceback (most recent call last):
File "D:\01_IAV-Codes\02_OCR-Models\test_01.py", line 24, in
frame = draw_ocr(frame, line, font_path=font_path)
File "C:\Users\I011786\Miniconda\envs\env-paddle\lib\site-packages\paddleocr\tools\infer\utility.py", line 384, in draw_ocr
scores = [1] * len(boxes)
TypeError: object of type 'NoneType' has no len()

[UserWangZz]

It seems that there is no target during the detection stage, which leads to an error in obtaining the length of the boxes. You can check whether the OCR detection model parameter configuration is correct, or whether the current frame contains text targets.