|
|
@@ -0,0 +1,178 @@
|
|
|
+# This Python file uses the following encoding: utf-8
|
|
|
+# @author runhey
|
|
|
+# github https://github.com/runhey
|
|
|
+import multiprocessing
|
|
|
+import pickle
|
|
|
+import socket
|
|
|
+import time
|
|
|
+from typing import Any, Dict, List, Optional
|
|
|
+
|
|
|
+import numpy as np
|
|
|
+import zerorpc
|
|
|
+
|
|
|
+from module.exception import ScriptError
|
|
|
+from module.logger import logger
|
|
|
+from module.ocr.ppocr import TextSystem
|
|
|
+
|
|
|
+_OCR_SERVER_PROCESS: Optional[multiprocessing.Process] = None
|
|
|
+
|
|
|
+
|
|
|
+def _normalize_address(address: str) -> str:
|
|
|
+ if address.startswith("tcp://"):
|
|
|
+ return address
|
|
|
+ return f"tcp://{address}"
|
|
|
+
|
|
|
+
|
|
|
+def _split_host_port(address: str) -> tuple[str, int]:
|
|
|
+ addr = address.replace("tcp://", "")
|
|
|
+ if ":" not in addr:
|
|
|
+ return addr, 22268
|
|
|
+ host, port = addr.rsplit(":", 1)
|
|
|
+ return host, int(port)
|
|
|
+
|
|
|
+
|
|
|
+def _is_port_in_use(host: str, port: int) -> bool:
|
|
|
+ s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
|
|
|
+ try:
|
|
|
+ s.settimeout(0.5)
|
|
|
+ s.connect((host, port))
|
|
|
+ s.shutdown(2)
|
|
|
+ return True
|
|
|
+ except Exception:
|
|
|
+ return False
|
|
|
+ finally:
|
|
|
+ s.close()
|
|
|
+
|
|
|
+
|
|
|
+def ensure_ocr_server_started() -> bool:
|
|
|
+ from module.server.setting import State
|
|
|
+
|
|
|
+ deploy_config = State.deploy_config
|
|
|
+ if not deploy_config.StartOcrServer:
|
|
|
+ return False
|
|
|
+
|
|
|
+ if deploy_config.OcrServerPort:
|
|
|
+ port = int(deploy_config.OcrServerPort)
|
|
|
+ else:
|
|
|
+ _, port = _split_host_port(str(deploy_config.OcrClientAddress))
|
|
|
+ host = "0.0.0.0"
|
|
|
+
|
|
|
+ if _is_port_in_use("127.0.0.1", port):
|
|
|
+ logger.info(f"OCR server already running on port {port}")
|
|
|
+ return True
|
|
|
+
|
|
|
+ global _OCR_SERVER_PROCESS
|
|
|
+ if _OCR_SERVER_PROCESS is not None and _OCR_SERVER_PROCESS.is_alive():
|
|
|
+ logger.info("OCR server process already started")
|
|
|
+ return True
|
|
|
+
|
|
|
+ _OCR_SERVER_PROCESS = multiprocessing.Process(
|
|
|
+ target=run_ocr_server,
|
|
|
+ args=(host, port),
|
|
|
+ name="ocr_server",
|
|
|
+ daemon=True,
|
|
|
+ )
|
|
|
+ _OCR_SERVER_PROCESS.start()
|
|
|
+ logger.info(f"Start OCR server on {host}:{port}")
|
|
|
+ for _ in range(50):
|
|
|
+ if _is_port_in_use("127.0.0.1", port):
|
|
|
+ return True
|
|
|
+ time.sleep(0.1)
|
|
|
+ logger.error(f"OCR server is not ready on port {port}")
|
|
|
+ return False
|
|
|
+
|
|
|
+
|
|
|
+def run_ocr_server(host: str, port: int) -> None:
|
|
|
+ server = zerorpc.Server(OcrServer())
|
|
|
+ server.bind(f"tcp://{host}:{port}")
|
|
|
+ server.run()
|
|
|
+
|
|
|
+
|
|
|
+class OcrServer:
|
|
|
+ def __init__(self) -> None:
|
|
|
+ self.model = TextSystem()
|
|
|
+
|
|
|
+ def ping(self) -> bool:
|
|
|
+ return True
|
|
|
+
|
|
|
+ @staticmethod
|
|
|
+ def _rotate_vertical(image: np.ndarray) -> np.ndarray:
|
|
|
+ height, width = image.shape[0:2]
|
|
|
+ if height * 1.0 / width >= 1.5:
|
|
|
+ return np.rot90(image)
|
|
|
+ return image
|
|
|
+
|
|
|
+ def ocr_single_line(self, image_bytes: bytes):
|
|
|
+ image = pickle.loads(image_bytes)
|
|
|
+ result, score = self.model.ocr_single_line(image)
|
|
|
+ return result, float(score)
|
|
|
+
|
|
|
+ def detect_and_ocr(
|
|
|
+ self,
|
|
|
+ image_bytes: bytes,
|
|
|
+ drop_score: float = 0.5,
|
|
|
+ unclip_ratio: Optional[float] = None,
|
|
|
+ box_thresh: Optional[float] = None,
|
|
|
+ vertical: bool = False,
|
|
|
+ ) -> List[Dict[str, Any]]:
|
|
|
+ image = pickle.loads(image_bytes)
|
|
|
+ if not vertical:
|
|
|
+ results = self.model.detect_and_ocr(image, drop_score=drop_score,
|
|
|
+ unclip_ratio=unclip_ratio,
|
|
|
+ box_thresh=box_thresh)
|
|
|
+ return [
|
|
|
+ {"box": r.box.tolist(), "ocr_text": r.ocr_text, "score": float(r.score)}
|
|
|
+ for r in results
|
|
|
+ ]
|
|
|
+
|
|
|
+ text_recognizer = self.model.text_recognizer
|
|
|
+
|
|
|
+ def vertical_text_recognizer(img_crop_list):
|
|
|
+ img_crop_list = [self._rotate_vertical(i) for i in img_crop_list]
|
|
|
+ return text_recognizer(img_crop_list)
|
|
|
+
|
|
|
+ self.model.text_recognizer = vertical_text_recognizer
|
|
|
+ try:
|
|
|
+ results = self.model.detect_and_ocr(image, drop_score=drop_score,
|
|
|
+ unclip_ratio=unclip_ratio,
|
|
|
+ box_thresh=box_thresh)
|
|
|
+ finally:
|
|
|
+ self.model.text_recognizer = text_recognizer
|
|
|
+
|
|
|
+ return [
|
|
|
+ {"box": r.box.tolist(), "ocr_text": r.ocr_text, "score": float(r.score)}
|
|
|
+ for r in results
|
|
|
+ ]
|
|
|
+
|
|
|
+
|
|
|
+class ModelProxy:
|
|
|
+ is_proxy = True
|
|
|
+
|
|
|
+ def __init__(self, address: str) -> None:
|
|
|
+ self.address = _normalize_address(address)
|
|
|
+ self.client = zerorpc.Client()
|
|
|
+ try:
|
|
|
+ self.client.connect(self.address)
|
|
|
+ self.client.ping()
|
|
|
+ except Exception as e:
|
|
|
+ raise ScriptError(f"OCR server connection failed: {self.address}") from e
|
|
|
+
|
|
|
+ def ocr_single_line(self, image: np.ndarray):
|
|
|
+ payload = pickle.dumps(image, protocol=4)
|
|
|
+ return self.client.ocr_single_line(payload)
|
|
|
+
|
|
|
+ def detect_and_ocr(
|
|
|
+ self,
|
|
|
+ image: np.ndarray,
|
|
|
+ drop_score: float = 0.5,
|
|
|
+ unclip_ratio: Optional[float] = None,
|
|
|
+ box_thresh: Optional[float] = None,
|
|
|
+ vertical: bool = False,
|
|
|
+ ):
|
|
|
+ payload = pickle.dumps(image, protocol=4)
|
|
|
+ results = self.client.detect_and_ocr(payload, drop_score, unclip_ratio, box_thresh, vertical)
|
|
|
+ from ppocronnx.predict_system import BoxedResult
|
|
|
+ return [
|
|
|
+ BoxedResult(np.array(item["box"]), None, item["ocr_text"], item["score"])
|
|
|
+ for item in results
|
|
|
+ ]
|