PPOCRLabel
- 使用PPOCRLabel对ocr预标注结果进行纠正
- 由于PaddleOCR代码库十分混乱,路径经常乱调pip和代码库的代码(pip库和源码冲突),经常报错,因此paddleocr和ppocrlabel都是使用pip包;
- PPOCRLabel中使用了cv2进行图片数据的读取,然cv2对中文路径读取有问题,经常会导致一些图片数据无法进行正常数据,在次基础上,把读取图片由cv2改为pilow,可以规避很多读取错误;
安装
pip install PPOCRLabel==2.1.3
启动
PPOCRLabel --lang ch
修改记录
修改1:
- 报错:AttributeError: ‘NoneType’ object has no attribute ‘shape’
'NoneType' object has no attribute 'shape'
Traceback (most recent call last):
File "D:\ProgramData\Anaconda3\envs\paddle\lib\site-packages\PPOCRLabel\libs\autoDialog.py", line 41, in run
h, w, _ = cv2.imdecode(np.fromfile(Imgpath, dtype=np.uint8), 1).shape
AttributeError: 'NoneType' object has no attribute 'shape'
- 原因:cv2不支持读取带有中文字符路径文件
h, w, _ = cv2.imdecode(np.fromfile(Imgpath, dtype=np.uint8), 1).shape
- 解决:使用PIL替换cv2, 将上述代码替换为
# 修改文件:"D:\ProgramData\Anaconda3\envs\paddle\lib\site-packages\PPOCRLabel\libs\autoDialog.py", line 41
# 增加PIL读取图片函数
from PIL import Image
def load_image(image_path: str, return_chw: bool = True, size: tuple = None):
image = Image.open(image_path).convert("RGB")
if size is not None:
image = image.resize(size) # resize image
image = np.asarray(image)
image = image[:, :, ::-1] # flip color channels from RGB to BGR
w, h = image.shape[1], image.shape[0] # update size after resize
if return_chw:
image = image.transpose(2, 0, 1)
return image, (w, h)
# 替换上面代码
try:
image_data, (w, h) = load_image(Imgpath, return_chw=False)
except Exception as e:
print(f"load file {Imgpath} fail!")
continue
修改2
- 报错:AttributeError: ‘NoneType’ object has no attribute ‘shape’
Traceback (most recent call last):
File "D:\ProgramData\Anaconda3\envs\paddle\lib\site-packages\PPOCRLabel\PPOCRLabel.py", line 1889, in saveFile
self._saveFile(imgidx, mode=mode)
File "D:\ProgramData\Anaconda3\envs\paddle\lib\site-packages\PPOCRLabel\PPOCRLabel.py", line 1934, in _saveFile
self.openNextImg()
File "D:\ProgramData\Anaconda3\envs\paddle\lib\site-packages\PPOCRLabel\PPOCRLabel.py", line 1880, in openNextImg
self.loadFile(filename)
File "D:\ProgramData\Anaconda3\envs\paddle\lib\site-packages\PPOCRLabel\PPOCRLabel.py", line 1550, in loadFile
height, width, depth = cvimg.shape
AttributeError: 'NoneType' object has no attribute 'shape'
- 原因还是cv2不能读取中文路径文件
- 解决:
# 修改: File "D:\ProgramData\Anaconda3\envs\paddle\lib\site-packages\PPOCRLabel\PPOCRLabel.py", line 1550, in loadFile
# cvimg = cv2.imdecode(np.fromfile(unicodeFilePath, dtype=np.uint8), 1)
cvimg, _ = load_image(unicodeFilePath, return_chw=False)
修改3
-
报错:error: (-215:Assertion failed) _src.total() > 0 in function ‘cv::warpPerspective’
- 报错描述:在对PPOCRLABEL的框进行重新识别是,发生如下报错:
Can not recognise the detection box in xxxx,png. Please change manually' unicodeFilePath is J:\data\mllm-data\xxxxxxxxx\wKh2CWERPJOAY2x-AAE62o598k0620.png OpenCV(4.2.0) C:\projects\opencv-python\opencv\modules\imgproc\src\imgwarp.cpp:3143: error: (-215:Assertion failed) _src.total() > 0 in function 'cv::warpPerspective'
- 原因是我们的ocr预标注数据Label.txt是使用PIL读取图片数据,调用ppocr进行生产的(并不是在PPOCRLabel工具内部生产的), 当我们修改数据框后, PPOCRLabel尝试再次使用cv2进行读取原图,此时由于cv2对路径较为敏感,经常会读取文件失败,才会出现如上情况
-
解决:
依旧是修改PPOCRLabel源码, 把cv2读取改为PIL读取,就不惯着cv2的臭毛病…
# 修改如下代码
def reRecognition(self):
#img = cv2.imdecode(np.fromfile(self.filePath,dtype=np.uint8),1)
img, _ = load_image(self.filePath, return_chw=False)
修改4
- 运行PPOCRLabel源码(paddleocr使用pip安装),报错: AttributeError: ‘Namespace’ object has no attribute ‘return_word_box’
- File "G:\dongyongfei786\paddle\PaddleOCR\ppstructure\predict_system.py", line 82, in __init__
self.return_word_box = args.return_word_box
AttributeError: 'Namespace' object has no attribute 'return_word_box'
- 原因:
- paddleocr使用pip安装的源码中(paddleocr=2.7.0.3), D:\ProgramData\Anaconda3\Lib\site-packages\paddleocr\tools\infer\utility.py, 缺少
# extended function
parser.add_argument("--return_word_box", type=str2bool, default=False, help='Whether return the bbox of each word (split by space) or chinese character. Only used in ppstructure for layout recovery')
赞赏
- 都说书中自有黄金屋,在这个“以钱为尊”的年代,没钱那可是万万不能,如果上述对各位帅哥美女有帮助的话,也可动一下发财的小手,你的支持的我做大的动力;
- 后续有需要考虑打成一个whl,供需要的小伙伴使用;
- 后续如果还有改动,会继续更新;