| from __future__ import annotations |
|
|
| import logging |
| import threading |
| from typing import TYPE_CHECKING, Optional |
|
|
| import cv2 |
|
|
| if TYPE_CHECKING: |
| import easyocr |
|
|
| logger = logging.getLogger("uvicorn") |
|
|
|
|
| def _load_reader(languages: list[str]) -> easyocr.Reader: |
| """Load the OCR model. easyocr and torch are imported here: they take seconds to import. |
| |
| Args: |
| languages (list[str]): The languages the reader must recognize. |
| |
| Raises: |
| OSError: When the model files can't be downloaded or read. |
| RuntimeError: When torch can't load the model files. |
| |
| Returns: |
| easyocr.Reader: The loaded reader. |
| """ |
| import easyocr |
| import torch |
|
|
| return easyocr.Reader(languages, gpu=torch.cuda.is_available()) |
|
|
|
|
| class OCRReader: |
| _instance: Optional["OCRReader"] = None |
| _lock = threading.Lock() |
| ocr_reader: easyocr.Reader |
|
|
| def __new__(cls): |
| """Return the one instance, loading the model on the first call. |
| |
| Raises: |
| OSError: When the model files can't be downloaded or read. |
| RuntimeError: When torch can't load the model files. |
| |
| Returns: |
| OCRReader: The shared instance, with its reader loaded. |
| """ |
| |
| |
| if cls._instance is None: |
| with cls._lock: |
| if cls._instance is None: |
| logger.info("Loading the OCR model into memory...") |
| instance = super().__new__(cls) |
| instance.ocr_reader = _load_reader(["en", "fr"]) |
| |
| |
| cls._instance = instance |
| return cls._instance |
|
|
| def read_text(self, img: cv2.typing.MatLike | None): |
| res = self.ocr_reader.readtext(img, detail=0) |
| if not isinstance(res, list): |
| return None |
| return " ".join([str(item) for item in res]) |
|
|
