📄 Awesome OCR multiple programing languages toolkits based on ONNX Runtime, OpenVINO, MNN, PaddlePaddle, TensorRT and PyTorch.
fix: fixed issue #555
SWHL32dad48ebbfa
diff --git a/python/demo.py b/python/demo.pyindex 6a0e4e1..200c2ed 100644--- a/python/demo.py+++ b/python/demo.py@@ -6,7 +6,8 @@ from rapidocr import RapidOCRengine = RapidOCR()img_url = "https://img1.baidu.com/it/u=3619974146,1266987475&fm=253&fmt=auto&app=138&f=JPEG?w=500&h=516"-result = engine(img_url)+img_url = "tests/test_files/return_word_debug.jpg"+result = engine(img_url, return_word_box=True)print(result)result.vis("vis_result.jpg")diff --git a/python/rapidocr/__init__.py b/python/rapidocr/__init__.pyindex a81fcfd..15e216c 100644--- a/python/rapidocr/__init__.py+++ b/python/rapidocr/__init__.py@@ -2,5 +2,6 @@# @Author: SWHL# @Contact: [email protected]from .main import RapidOCR-from .utils import LoadImageError, VisRes+from .utils.load_image import LoadImageErrorfrom .utils.typings import EngineType, LangCls, LangDet, LangRec, ModelType, OCRVersion+from .utils.vis_res import VisResdiff --git a/python/rapidocr/cal_rec_boxes/main.py b/python/rapidocr/cal_rec_boxes/main.pyindex 928312c..2b8728f 100644--- a/python/rapidocr/cal_rec_boxes/main.py+++ b/python/rapidocr/cal_rec_boxes/main.py@@ -25,13 +25,13 @@ class CalRecBoxes:def __call__(self,imgs: List[np.ndarray],- dt_boxes: List[np.ndarray],+ dt_boxes: np.ndarray,rec_res: TextRecOutput,return_single_char_box: bool = False,) -> TextRecOutput:word_results = []for idx, (img, box) in enumerate(zip(imgs, dt_boxes)):- if rec_res.txts is None:+ if rec_res.txts is None or img.size == 0 or rec_res.word_results is None:continueh, w = img.shape[:2]diff --git a/python/rapidocr/ch_ppocr_det/main.py b/python/rapidocr/ch_ppocr_det/main.pyindex 146f3d5..9c2736f 100644--- a/python/rapidocr/ch_ppocr_det/main.py+++ b/python/rapidocr/ch_ppocr_det/main.py@@ -15,7 +15,7 @@# @Author: SWHL# @Contact: [email protected]import time-from typing import Any, Dict, List+from typing import Any, Dictimport numpy as np@@ -77,7 +77,7 @@ class TextDetector:return DetPreProcess(limit_side_len, self.limit_type, self.mean, self.std)@staticmethod- def sorted_boxes(dt_boxes: np.ndarray) -> List[np.ndarray]:+ def sorted_boxes(dt_boxes: np.ndarray) -> np.ndarray:"""Sort text boxes in order from top to bottom, left to rightargs:@@ -100,4 +100,4 @@ class TextDetector:_boxes[j + 1] = tmpelse:break- return _boxes+ return np.array(_boxes)diff --git a/python/rapidocr/ch_ppocr_det/utils.py b/python/rapidocr/ch_ppocr_det/utils.pyindex 1311f32..8449d81 100644--- a/python/rapidocr/ch_ppocr_det/utils.py+++ b/python/rapidocr/ch_ppocr_det/utils.py@@ -18,7 +18,7 @@ from ..utils.vis_res import VisResclass TextDetOutput:img: Optional[np.ndarray] = Noneboxes: Optional[np.ndarray] = None- scores: Optional[Tuple[float]] = None+ scores: Optional[List[float]] = Noneelapse: float = 0.0def __len__(self):diff --git a/python/rapidocr/ch_ppocr_rec/main.py b/python/rapidocr/ch_ppocr_rec/main.pyindex 430174a..43789f5 100644--- a/python/rapidocr/ch_ppocr_rec/main.py+++ b/python/rapidocr/ch_ppocr_rec/main.py@@ -35,7 +35,6 @@ DEFAULT_MODEL_PATH = Path(__file__).parent.parent / "models"class TextRecognizer:def __init__(self, cfg: Dict[str, Any]):self.session = get_engine(cfg.engine_type)(cfg)- self.logger = logger# onnx has inner character, other engine get or download character_dict_pathcharacter, character_dict_path = self.get_character_dict(cfg)@@ -77,7 +76,7 @@ class TextRecognizer:file_url=dict_download_url,sha256=None,save_path=dict_path,- logger=self.logger,+ logger=logger,))diff --git a/python/rapidocr/ch_ppocr_rec/typings.py b/python/rapidocr/ch_ppocr_rec/typings.pyindex 0239d9f..a61952a 100644--- a/python/rapidocr/ch_ppocr_rec/typings.py+++ b/python/rapidocr/ch_ppocr_rec/typings.py@@ -36,7 +36,7 @@ class TextRecInput:class TextRecOutput:imgs: Optional[List[np.ndarray]] = Nonetxts: Optional[Tuple[str]] = None- scores: Tuple[float] = (1.0,)+ scores: List[float] = field(default_factory=lambda: [1.0])word_results: Tuple[Tuple[str, float, Optional[List[List[int]]]]] = (("", 1.0, None),)diff --git a/python/rapidocr/inference_engine/onnxruntime/main.py b/python/rapidocr/inference_engine/onnxruntime/main.pyindex 411eacf..7ed03b7 100644--- a/python/rapidocr/inference_engine/onnxruntime/main.py+++ b/python/rapidocr/inference_engine/onnxruntime/main.py@@ -9,7 +9,7 @@ from typing import Any, Dict, Listimport numpy as npfrom onnxruntime import GraphOptimizationLevel, InferenceSession, SessionOptions-from ...utils import DownloadFile, DownloadFileInput+from ...utils.download_file import DownloadFile, DownloadFileInputfrom ...utils.log import loggerfrom ..base import FileInfo, InferSessionfrom .provider_config import ProviderConfig@@ -17,8 +17,6 @@ from .provider_config import ProviderConfigclass OrtInferSession(InferSession):def __init__(self, cfg: Dict[str, Any]):- self.logger = logger-# support custom session (PR #451)session = cfg.get("session", None)if session is not None:@@ -27,7 +25,7 @@ class OrtInferSession(InferSession):f"Expected session to be an InferenceSession, got {type(session)}")- self.logger.debug("Using the provided InferenceSession for inference.")+ logger.debug("Using the provided InferenceSession for inference.")self.session = sessionreturn@@ -48,11 +46,11 @@ class OrtInferSession(InferSession):file_url=model_info["model_dir"],sha256=model_info["SHA256"],save_path=model_path,- logger=self.logger,+ logger=logger,)DownloadFile.run(download_params)- self.logger.info(f"Using {model_path}")+ logger.info(f"Using {model_path}")model_path = Path(model_path)self._verify_model(model_path)diff --git a/python/rapidocr/inference_engine/onnxruntime/provider_config.py b/python/rapidocr/inference_engine/onnxruntime/provider_config.pyindex 8b47144..deff27a 100644--- a/python/rapidocr/inference_engine/onnxruntime/provider_config.py+++ b/python/rapidocr/inference_engine/onnxruntime/provider_config.py@@ -19,8 +19,6 @@ class EP(Enum):class ProviderConfig:def __init__(self, engine_cfg: Dict[str, Any]):- self.logger = logger-self.had_providers: List[str] = get_available_providers()self.default_provider = self.had_providers[0]@@ -37,13 +35,13 @@ class ProviderConfig:results.insert(0, (EP.CUDA_EP.value, self.cuda_ep_cfg()))if self.is_dml_available():- self.logger.info(+ logger.info("Windows 10 or above detected, try to use DirectML as primary provider")results.insert(0, (EP.DIRECTML_EP.value, self.dml_ep_cfg()))if self.is_cann_available():- self.logger.info("Try to use CANNExecutionProvider to infer")+ logger.info("Try to use CANNExecutionProvider to infer")results.insert(0, (EP.CANN_EP.value, self.cann_ep_cfg()))return results@@ -79,10 +77,10 @@ class ProviderConfig:for ep, check_func in providers_to_check.items():if check_func() and first_provider != ep.value:- self.logger.warning(+ logger.warning(f"{ep.value} is available, but the inference part is automatically shifted to be executed under {first_provider}. ")- self.logger.warning(f"The available lists are {session_providers}")+ logger.warning(f"The available lists are {session_providers}")def is_cuda_available(self) -> bool:if not self.cfg_use_cuda:@@ -92,7 +90,7 @@ class ProviderConfig:if get_device() == "GPU" and CUDA_EP in self.had_providers:return True- self.logger.warning(+ logger.warning(f"{CUDA_EP} is not in available providers ({self.had_providers}). Use {self.default_provider} inference by default.")install_instructions = [@@ -113,7 +111,7 @@ class ProviderConfig:cur_os = platform.system()if cur_os != "Windows":- self.logger.warning(+ logger.warning(f"DirectML is only supported in Windows OS. The current OS is {cur_os}. Use {self.default_provider} inference by default.",)return False@@ -123,7 +121,7 @@ class ProviderConfig:int(window_build_number_str) if window_build_number_str.isdigit() else 0)if window_build_number < 18362:- self.logger.warning(+ logger.warning(f"DirectML is only supported in Windows 10 Build 18362 and above OS. The current Windows Build is {window_build_number}. Use {self.default_provider} inference by default.",)return False@@ -132,7 +130,7 @@ class ProviderConfig:if DML_EP in self.had_providers:return True- self.logger.warning(+ logger.warning(f"{DML_EP} is not in available providers ({self.had_providers}). Use {self.default_provider} inference by default.")install_instructions = [@@ -152,7 +150,7 @@ class ProviderConfig:if CANN_EP in self.had_providers:return True- self.logger.warning(+ logger.warning(f"{CANN_EP} is not in available providers ({self.had_providers}). Use {self.default_provider} inference by default.")install_instructions = [@@ -167,4 +165,4 @@ class ProviderConfig:def print_log(self, log_list: List[str]):for log_info in log_list:- self.logger.info(log_info)+ logger.info(log_info)diff --git a/python/rapidocr/inference_engine/openvino.py b/python/rapidocr/inference_engine/openvino.pyindex a9f0484..11b905c 100644--- a/python/rapidocr/inference_engine/openvino.py+++ b/python/rapidocr/inference_engine/openvino.py@@ -18,7 +18,6 @@ from .base import FileInfo, InferSessionclass OpenVINOInferSession(InferSession):def __init__(self, cfg: DictConfig):super().__init__(cfg)- self.logger = loggercore = Core()@@ -38,11 +37,11 @@ class OpenVINOInferSession(InferSession):file_url=model_info["model_dir"],sha256=model_info["SHA256"],save_path=model_path,- logger=self.logger,+ logger=logger,)DownloadFile.run(download_params)- self.logger.info(f"Using {model_path}")+ logger.info(f"Using {model_path}")model_path = Path(model_path)self._verify_model(model_path)@@ -85,7 +84,7 @@ class OpenVINOInferSession(InferSession):if scheduling_core_type is not None:config["SCHEDULING_CORE_TYPE"] = str(scheduling_core_type)- self.logger.info(f"Using OpenVINO config: {config}")+ logger.info(f"Using OpenVINO config: {config}")return configdef __call__(self, input_content: np.ndarray) -> np.ndarray:diff --git a/python/rapidocr/inference_engine/paddle.py b/python/rapidocr/inference_engine/paddle.pyindex c8a5679..5ae6008 100644--- a/python/rapidocr/inference_engine/paddle.py+++ b/python/rapidocr/inference_engine/paddle.py@@ -19,7 +19,6 @@ from .base import FileInfo, InferSessionclass PaddleInferSession(InferSession):def __init__(self, cfg, mode: Optional[str] = None) -> None:- self.logger = loggerself.mode = mode# 设置设备特定的环境变量@@ -37,7 +36,7 @@ class PaddleInferSession(InferSession):def _set_envs(envs):for key, val in envs.items():os.environ[key] = val- self.logger.debug(f"{key} has been set to {val}.")+ logger.debug(f"{key} has been set to {val}.")if engine_cfg.get("use_cuda", False):envs = {}@@ -97,8 +96,8 @@ class PaddleInferSession(InferSession):model_info, default_model_dir, pdiparams_name)- self.logger.info(f"Using {pdmodel_path}")- self.logger.info(f"Using {pdiparams_path}")+ logger.info(f"Using {pdmodel_path}")+ logger.info(f"Using {pdiparams_path}")return pdmodel_path, pdiparams_pathmodel_dir = Path(model_dir)@@ -109,8 +108,8 @@ class PaddleInferSession(InferSession):self._verify_model(pdmodel_path)self._verify_model(pdiparams_path)- self.logger.info(f"Using {pdmodel_path}")- self.logger.info(f"Using {pdiparams_path}")+ logger.info(f"Using {pdmodel_path}")+ logger.info(f"Using {pdiparams_path}")return pdmodel_path, pdiparams_pathdef download_model(@@ -125,7 +124,7 @@ class PaddleInferSession(InferSession):file_url=model_file_url,sha256=model_info[model_file_name],save_path=model_file_path,- logger=self.logger,+ logger=logger,))return model_file_path@@ -136,27 +135,27 @@ class PaddleInferSession(InferSession):if cfg.engine_cfg.use_cuda:gpu_id = self.get_infer_gpuid()if gpu_id is None:- self.logger.warning(+ logger.warning("GPU is not found in current device by nvidia-smi. Please check your device or ignore it if run on jetson.")device_id = cfg.engine_cfg.gpu_idinfer_opts.enable_use_gpu(cfg.engine_cfg.gpu_mem, device_id)- self.logger.info(f"Using GPU device with ID: {device_id}")+ logger.info(f"Using GPU device with ID: {device_id}")elif cfg.engine_cfg.use_npu:npu_id = cfg.engine_cfg.get("npu_id", 0)infer_opts.enable_custom_device("npu", npu_id)- self.logger.info(f"Using NPU device with ID: {npu_id}")+ logger.info(f"Using NPU device with ID: {npu_id}")else: # CPUinfer_opts.disable_gpu()- self.logger.info("Using CPU device")+ logger.info("Using CPU device")cpu_nums = os.cpu_count()infer_num_threads = cfg.engine_cfg.get("cpu_math_library_num_threads", -1)if infer_num_threads != -1 and 1 <= infer_num_threads <= cpu_nums:infer_opts.set_cpu_math_library_num_threads(infer_num_threads)- self.logger.info(f"Set CPU math library threads to: {infer_num_threads}")+ logger.info(f"Set CPU math library threads to: {infer_num_threads}")# enable memory optiminfer_opts.enable_memory_optim()@@ -174,31 +173,29 @@ class PaddleInferSession(InferSession):if cfg.engine_cfg.use_cuda:gpu_id = self.get_infer_gpuid()if gpu_id is None:- self.logger.warning(+ logger.warning("GPU is not found in current device by nvidia-smi. Please check your device or ignore it if run on jetson.")device_id = cfg.engine_cfg.gpu_idinfer_opts.enable_use_gpu(cfg.engine_cfg.gpu_mem, device_id)- self.logger.info(f"Using GPU device with ID: {device_id}")+ logger.info(f"Using GPU device with ID: {device_id}")elif cfg.engine_cfg.use_npu:npu_id = cfg.engine_cfg.get("npu_id", 0)infer_opts.enable_custom_device("npu", npu_id)- self.logger.info(f"Using NPU device with ID: {npu_id}")+ logger.info(f"Using NPU device with ID: {npu_id}")else: # CPUinfer_opts.disable_gpu()if hasattr(infer_opts, "disable_mkldnn"):infer_opts.disable_mkldnn()- self.logger.info("Using CPU device")+ logger.info("Using CPU device")cpu_nums = os.cpu_count()infer_num_threads = cfg.engine_cfg.get("cpu_math_library_num_threads", -1)if infer_num_threads != -1 and 1 <= infer_num_threads <= cpu_nums:infer_opts.set_cpu_math_library_num_threads(infer_num_threads)- self.logger.info(- f"Set CPU math library threads to: {infer_num_threads}"- )+ logger.info(f"Set CPU math library threads to: {infer_num_threads}")if hasattr(infer_opts, "enable_new_ir"):infer_opts.enable_new_ir(True)diff --git a/python/rapidocr/main.py b/python/rapidocr/main.pyindex b880368..544678f 100644--- a/python/rapidocr/main.py+++ b/python/rapidocr/main.py@@ -15,18 +15,18 @@ from .ch_ppocr_cls import TextClassifier, TextClsOutputfrom .ch_ppocr_det import TextDetector, TextDetOutputfrom .ch_ppocr_rec import TextRecInput, TextRecognizer, TextRecOutputfrom .cli import check_install, generate_cfg-from .utils import (- LoadImage,- RapidOCROutput,- VisRes,- add_round_letterbox,- get_padding_h,+from .utils.load_image import LoadImage+from .utils.log import logger+from .utils.output import RapidOCROutput+from .utils.parse_parameters import ParseParams+from .utils.process_img import (+ apply_vertical_padding,get_rotate_crop_image,+ map_boxes_to_original,resize_image_within_bounds,)-from .utils.log import logger-from .utils.parse_parameters import ParseParamsfrom .utils.typings import LangRec+from .utils.vis_res import VisResroot_dir = Path(__file__).resolve().parentDEFAULT_CFG_PATH = root_dir / "config.yaml"@@ -39,7 +39,6 @@ class RapidOCR:cfg = self._load_config(config_path, params)logger.setLevel(cfg.Global.log_level.upper())- self.logger = loggerself._initialize(cfg)@@ -109,64 +108,134 @@ class RapidOCR:ori_img = self.load_img(img_content)img, op_record = self.preprocess_img(ori_img)+ det_res, cls_res, rec_res, cropped_img_list = self.run_ocr_steps(img, op_record)+ return self.build_final_output(+ ori_img, det_res, cls_res, rec_res, cropped_img_list, op_record+ )+ def run_ocr_steps(self, img: np.ndarray, op_record: Dict[str, Any]):det_res, cls_res, rec_res = TextDetOutput(), TextClsOutput(), TextRecOutput()+if self.use_det:try:- img, det_res = self.get_det_res(img, op_record)+ cropped_img_list, det_res = self.detect_and_crop(img, op_record)except RapidOCRError as e:- self.logger.warning(e)- return RapidOCROutput()+ logger.warning(e)+ return TextDetOutput(), TextClsOutput(), TextRecOutput(), []+ else:+ cropped_img_list = [img]if self.use_cls:try:- img, cls_res = self.get_cls_res(img)+ cls_img_list, cls_res = self.cls_and_rotate(cropped_img_list)except RapidOCRError as e:- self.logger.warning(e)- return RapidOCROutput()+ logger.warning(e)+ return det_res, TextClsOutput(), TextRecOutput(), []+ else:+ cls_img_list = cropped_img_listif self.use_rec:- rec_res = self.get_rec_res(img)+ try:+ rec_res = self.recognize_txt(cls_img_list)+ except RapidOCRError as e:+ logger.warning(e)+ return det_res, cls_res, TextRecOutput(), []- return self.finalize_results(ori_img, det_res, cls_res, rec_res, img, op_record)+ return det_res, cls_res, rec_res, cropped_img_list- def finalize_results(+ def build_final_output(self,ori_img: np.ndarray,det_res: TextDetOutput,cls_res: TextClsOutput,rec_res: TextRecOutput,- img: List[np.ndarray],+ cropped_img_list: List[np.ndarray],op_record: Dict[str, Any],) -> Union[TextDetOutput, TextClsOutput, TextRecOutput, RapidOCROutput]:- raw_h, raw_w = ori_img.shape[:2]+ ori_h, ori_w = ori_img.shape[:2]++ if det_res.boxes is not None:+ det_res.boxes = map_boxes_to_original(+ det_res.boxes, op_record, ori_h, ori_w+ )++ # 过滤识别结果为空的值+ if (+ rec_res.txts is not None+ and det_res.boxes is not None+ and det_res.scores is not None+ ):+ empty_ids = {i for i, v in enumerate(rec_res.txts) if not v.strip()}++ det_res.boxes = np.array(+ [v for i, v in enumerate(det_res.boxes) if i not in empty_ids]+ )+ det_res.scores = [+ v for i, v in enumerate(det_res.scores) if i not in empty_ids+ ]++ rec_res.txts = [v for i, v in enumerate(rec_res.txts) if i not in empty_ids]+ rec_res.word_results = [+ v for i, v in enumerate(rec_res.word_results) if i not in empty_ids+ ]++ # 仅分类结果+ if (+ det_res.boxes is None+ and rec_res.txts is None+ and cls_res.cls_res is not None+ ):+ return cls_res++ # 无有效输出+ if det_res.boxes is None and rec_res.txts is None:+ return RapidOCROutput()++ # 仅识别结果(无检测)+ if det_res.boxes is None and rec_res.txts is not None:+ return rec_res++ # 仅检测结果(无识别)+ if det_res.boxes is not None and rec_res.txts is None:+ return det_resif (self.return_word_boxand det_res.boxes is not None- and all(v for v in rec_res.word_results)+ and all(rec_res.word_results)):rec_res.word_results = self.calc_word_boxes(- img, det_res, rec_res, op_record, raw_h, raw_w+ cropped_img_list, det_res.boxes, rec_res, op_record, ori_h, ori_w)- if det_res.boxes is not None:- det_res.boxes = self._get_origin_points(- det_res.boxes, op_record, raw_h, raw_w- )- return self.get_final_res(ori_img, det_res, cls_res, rec_res)+ ocr_res = RapidOCROutput(+ img=ori_img,+ boxes=det_res.boxes,+ txts=rec_res.txts,+ scores=rec_res.scores,+ word_results=rec_res.word_results,+ elapse_list=[det_res.elapse, cls_res.elapse, rec_res.elapse],+ viser=VisRes(+ text_score=self.cfg.Global.text_score,+ lang_type=self.cfg.Rec.lang_type,+ font_path=self.cfg.Global.font_path,+ ),+ )++ ocr_res = self.filter_by_text_score(ocr_res)+ return ocr_res if len(ocr_res) > 0 else RapidOCROutput()def calc_word_boxes(self,img: List[np.ndarray],- det_res: TextDetOutput,+ dt_boxes: np.ndarray,rec_res: TextRecOutput,op_record: Dict[str, Any],raw_h: int,raw_w: int,) -> Any:rec_res = self.cal_rec_boxes(- img, det_res.boxes, rec_res, self.return_single_char_box+ img, dt_boxes, rec_res, self.return_single_char_box)origin_words = []@@ -176,8 +245,8 @@ class RapidOCR:if bbox is None:continue- origin_words_points = self._get_origin_points(- [bbox], op_record, raw_h, raw_w+ origin_words_points = map_boxes_to_original(+ np.array([bbox]).astype(np.float64), op_record, raw_h, raw_w)origin_words_points = origin_words_points.astype(np.int32).tolist()[0]origin_words_item.append((txt, score, origin_words_points))@@ -188,14 +257,14 @@ class RapidOCR:def update_params(self,- use_det,- use_cls,- use_rec,- return_word_box,- return_single_char_box,- text_score,- box_thresh,- unclip_ratio,+ use_det: Optional[bool] = None,+ use_cls: Optional[bool] = None,+ use_rec: Optional[bool] = None,+ return_word_box: bool = False,+ return_single_char_box: bool = False,+ text_score: float = 0.5,+ box_thresh: float = 0.5,+ unclip_ratio: float = 1.6,):self.use_det = self.use_det if use_det is None else use_detself.use_cls = self.use_cls if use_cls is None else use_cls@@ -215,28 +284,30 @@ class RapidOCR:op_record["preprocess"] = {"ratio_h": ratio_h, "ratio_w": ratio_w}return img, op_record- def get_det_res(+ def detect_and_crop(self, img: np.ndarray, op_record: Dict[str, Any]) -> Tuple[List[np.ndarray], TextDetOutput]:- img, op_record = self._add_letterbox(img, op_record)+ img, op_record = apply_vertical_padding(+ img, op_record, self.width_height_ratio, self.min_height+ )det_res = self.text_det(img)+if det_res.boxes is None:raise RapidOCRError("The text detection result is empty")- img_list = self.get_crop_img_list(img, det_res)- return img_list, det_res+ img_crop_list = self.crop_text_regions(img, det_res.boxes)+ return img_crop_list, det_res- def get_crop_img_list(- self, img: np.ndarray, det_res: TextDetOutput+ def crop_text_regions(+ self, img: np.ndarray, det_boxes: np.ndarray) -> List[np.ndarray]:img_crop_list = []- for box in det_res.boxes:- tmp_box = copy.deepcopy(box)- img_crop = get_rotate_crop_image(img, tmp_box)+ for box in det_boxes:+ img_crop = get_rotate_crop_image(img, copy.deepcopy(box))img_crop_list.append(img_crop)return img_crop_list- def get_cls_res(+ def cls_and_rotate(self, img: List[np.ndarray]) -> Tuple[List[np.ndarray], TextClsOutput]:cls_res = self.text_cls(img)@@ -244,112 +315,36 @@ class RapidOCR:raise RapidOCRError("The text classifier is empty")return cls_res.img_list, cls_res- def get_rec_res(self, img: List[np.ndarray]) -> TextRecOutput:+ def recognize_txt(self, img: List[np.ndarray]) -> TextRecOutput:rec_input = TextRecInput(img=img, return_word_box=self.return_word_box)- return self.text_rec(rec_input)-- def get_final_res(- self,- ori_img: np.ndarray,- det_res: TextDetOutput,- cls_res: TextClsOutput,- rec_res: TextRecOutput,- ) -> Union[TextDetOutput, TextClsOutput, TextRecOutput, RapidOCROutput]:- dt_boxes = det_res.boxes- txt_res = rec_res.txts- if dt_boxes is None and txt_res is None and cls_res.cls_res is not None:- return cls_res-- if dt_boxes is None and txt_res is None:- return RapidOCROutput()+ rec_res = self.text_rec(rec_input)+ if rec_res.txts is None:+ raise RapidOCRError("The text recognize result is empty")- if dt_boxes is None and txt_res is not None:- return rec_res+ return rec_res- if dt_boxes is not None and txt_res is None:- return det_res-- ocr_res = RapidOCROutput(- img=ori_img,- boxes=det_res.boxes,- txts=rec_res.txts,- scores=rec_res.scores,- word_results=rec_res.word_results,- elapse_list=[det_res.elapse, cls_res.elapse, rec_res.elapse],- viser=VisRes(- text_score=self.cfg.Global.text_score,- lang_type=self.cfg.Rec.lang_type,- font_path=self.cfg.Global.font_path,- ),- )- ocr_res = self.filter_by_text_score(ocr_res)- if len(ocr_res) <= 0:- return RapidOCROutput()+ def filter_by_text_score(self, ocr_res: RapidOCROutput) -> RapidOCROutput:+ filter_boxes, filter_txts, filter_scores, filter_words = [], [], [], []+ for i, (box, txt, score) in enumerate(+ zip(ocr_res.boxes, ocr_res.txts, ocr_res.scores)+ ):+ if score < self.text_score:+ continue- return ocr_res+ if ocr_res.word_results[i]:+ filter_words.append(ocr_res.word_results[i])- def filter_by_text_score(self, ocr_res: RapidOCROutput) -> RapidOCROutput:- filter_boxes, filter_txts, filter_scores = [], [], []- for box, txt, score in zip(ocr_res.boxes, ocr_res.txts, ocr_res.scores):- if float(score) >= self.text_score:- filter_boxes.append(box)- filter_txts.append(txt)- filter_scores.append(score)+ filter_boxes.append(box)+ filter_txts.append(txt)+ filter_scores.append(score)ocr_res.boxes = np.array(filter_boxes)ocr_res.txts = tuple(filter_txts)ocr_res.scores = tuple(filter_scores)+ ocr_res.word_results = tuple(filter_words)return ocr_res- def _add_letterbox(- self, img: np.ndarray, op_record: Dict[str, Any]- ) -> Tuple[np.ndarray, Dict[str, Any]]:- h, w = img.shape[:2]-- if self.width_height_ratio == -1:- use_limit_ratio = False- else:- use_limit_ratio = w / h > self.width_height_ratio-- if h <= self.min_height or use_limit_ratio:- padding_h = get_padding_h(h, w, self.width_height_ratio, self.min_height)- block_img = add_round_letterbox(img, (padding_h, padding_h, 0, 0))- op_record["padding_1"] = {"top": padding_h, "left": 0}- return block_img, op_record-- op_record["padding_1"] = {"top": 0, "left": 0}- return img, op_record-- def _get_origin_points(- self,- dt_boxes: List[np.ndarray],- op_record: Dict[str, Any],- raw_h: int,- raw_w: int,- ) -> np.ndarray:- dt_boxes_array = np.array(dt_boxes).astype(np.float32)- for op in reversed(list(op_record.keys())):- v = op_record[op]- if "padding" in op:- top, left = v.get("top"), v.get("left")- dt_boxes_array[:, :, 0] -= left- dt_boxes_array[:, :, 1] -= top- elif "preprocess" in op:- ratio_h = v.get("ratio_h")- ratio_w = v.get("ratio_w")- dt_boxes_array[:, :, 0] *= ratio_w- dt_boxes_array[:, :, 1] *= ratio_h-- dt_boxes_array = np.where(dt_boxes_array < 0, 0, dt_boxes_array)- dt_boxes_array[..., 0] = np.where(- dt_boxes_array[..., 0] > raw_w, raw_w, dt_boxes_array[..., 0]- )- dt_boxes_array[..., 1] = np.where(- dt_boxes_array[..., 1] > raw_h, raw_h, dt_boxes_array[..., 1]- )- return dt_boxes_array-class RapidOCRError(Exception):passdiff --git a/python/rapidocr/utils/__init__.py b/python/rapidocr/utils/__init__.pyindex ff42bbd..0ecdd4f 100644--- a/python/rapidocr/utils/__init__.py+++ b/python/rapidocr/utils/__init__.py@@ -1,14 +1,3 @@# -*- encoding: utf-8 -*-# @Author: SWHL# @Contact: [email protected]-from .download_file import DownloadFile, DownloadFileException, DownloadFileInput-from .load_image import LoadImage, LoadImageError-from .output import RapidOCROutput-from .parse_parameters import ParseParams-from .process_img import (- add_round_letterbox,- get_padding_h,- get_rotate_crop_image,- resize_image_within_bounds,-)-from .vis_res import VisResdiff --git a/python/rapidocr/utils/process_img.py b/python/rapidocr/utils/process_img.pyindex 4d71249..46e8a16 100644--- a/python/rapidocr/utils/process_img.py+++ b/python/rapidocr/utils/process_img.py@@ -1,12 +1,56 @@# -*- encoding: utf-8 -*-# @Author: SWHL# @Contact: [email protected]-from typing import Tuple+from typing import Any, Dict, Tupleimport cv2import numpy as np+def map_boxes_to_original(+ dt_boxes: np.ndarray, op_record: Dict[str, Any], ori_h: int, ori_w: int+) -> np.ndarray:+ for op in reversed(list(op_record.keys())):+ v = op_record[op]+ if "padding" in op:+ top, left = v.get("top"), v.get("left")+ dt_boxes[:, :, 0] -= left+ dt_boxes[:, :, 1] -= top+ elif "preprocess" in op:+ ratio_h = v.get("ratio_h")+ ratio_w = v.get("ratio_w")+ dt_boxes[:, :, 0] *= ratio_w+ dt_boxes[:, :, 1] *= ratio_h++ dt_boxes = np.where(dt_boxes < 0, 0, dt_boxes)+ dt_boxes[..., 0] = np.where(dt_boxes[..., 0] > ori_w, ori_w, dt_boxes[..., 0])+ dt_boxes[..., 1] = np.where(dt_boxes[..., 1] > ori_h, ori_h, dt_boxes[..., 1])+ return dt_boxes+++def apply_vertical_padding(+ img: np.ndarray,+ op_record: Dict[str, Any],+ width_height_ratio: float,+ min_height: float,+) -> Tuple[np.ndarray, Dict[str, Any]]:+ h, w = img.shape[:2]++ if width_height_ratio == -1:+ use_limit_ratio = False+ else:+ use_limit_ratio = w / h > width_height_ratio++ if h <= min_height or use_limit_ratio:+ padding_h = get_padding_h(h, w, width_height_ratio, min_height)+ block_img = add_round_letterbox(img, (padding_h, padding_h, 0, 0))+ op_record["padding_1"] = {"top": padding_h, "left": 0}+ return block_img, op_record++ op_record["padding_1"] = {"top": 0, "left": 0}+ return img, op_record++def get_padding_h(h: int, w: int, width_height_ratio: float, min_height: float) -> int:new_h = max(int(w / width_height_ratio), min_height) * 2padding_h = int(abs(new_h - h) / 2)diff --git a/python/rapidocr/utils/vis_res.py b/python/rapidocr/utils/vis_res.pyindex f6f3c22..f228897 100644--- a/python/rapidocr/utils/vis_res.py+++ b/python/rapidocr/utils/vis_res.py@@ -31,8 +31,6 @@ class VisRes:lang_type: Optional[LangRec] = None,font_path: Optional[str] = None,):- self.logger = logger-self.text_score = text_scoreself.load_img = LoadImage()@@ -44,9 +42,9 @@ class VisRes:img_content: InputType,dt_boxes: np.ndarray,txts: Optional[Union[List[str], Tuple[str]]] = None,- scores: Optional[Tuple[float]] = None,+ scores: Optional[List[float]] = None,) -> np.ndarray:- self.logger.info(f"Using {self.font_path} to visualize results.")+ logger.info(f"Using {self.font_path} to visualize results.")if txts is None:return self.draw_dt_boxes(img_content, dt_boxes, scores)@@ -56,7 +54,7 @@ class VisRes:self,img_content: InputType,dt_boxes: np.ndarray,- scores: Optional[Tuple[float]] = None,+ scores: Optional[List[float]] = None,) -> np.ndarray:img = self.load_img(img_content)if scores is None:@@ -90,7 +88,7 @@ class VisRes:file_url=default_info["path"],sha256=default_info["SHA256"],save_path=DEFAULT_FONT_PATH,- logger=self.logger,+ logger=logger,)if lang_type is None:@@ -106,7 +104,7 @@ class VisRes:font_url, font_sha256 = font_info["path"], font_info["SHA256"]if font_url is None:- self.logger.warning(+ logger.warning("Font file for %s is not found in the supported font list. Default font file will be used.",lang_type,)@@ -119,7 +117,7 @@ class VisRes:file_url=font_url,sha256=font_sha256,save_path=save_font_path,- logger=self.logger,+ logger=logger,verbose=False,)DownloadFile.run(input_param)@@ -131,7 +129,7 @@ class VisRes:self,imgs: Sequence[InputType],txts: Union[List[str], Tuple[str]],- scores: Tuple[float],+ scores: List[float],) -> np.ndarray:result_imgs = []for img, txt, score in zip(imgs, txts, scores):@@ -178,7 +176,7 @@ class VisRes:img_content: InputType,dt_boxes: np.ndarray,txts: Union[List[str], Tuple[str]],- scores: Optional[Tuple[float]] = None,+ scores: Optional[List[float]] = None,) -> np.ndarray:image = Image.fromarray(self.load_img(img_content))h, w = image.height, image.widthdiff --git a/python/tests/test_files/check_return_word_len.jpeg b/python/tests/test_files/check_return_word_len.jpegnew file mode 100644index 0000000..583553cBinary files /dev/null and b/python/tests/test_files/check_return_word_len.jpeg differdiff --git a/python/tests/test_return_word.py b/python/tests/test_return_word.pyindex 0814c5e..1c8cc84 100644--- a/python/tests/test_return_word.py+++ b/python/tests/test_return_word.py@@ -13,7 +13,7 @@ sys.path.append(str(root_dir))from rapidocr import RapidOCR-tests_dir = root_dir / "tests" / "test_files"+test_dir = root_dir / "tests" / "test_files"@pytest.fixture()@@ -22,6 +22,13 @@ def engine():return engine+def test_txts_equal_words(engine):+ img_path = test_dir / "check_return_word_len.jpeg"+ result = engine(img_path, return_word_box=True)++ assert len(result.txts) == len(result.word_results)++@mark.parametrize("img_name,words",[@@ -36,7 +43,7 @@ def engine():],)def test_cn_word_ocr(engine, img_name: str, words: List[str]):- img_path = tests_dir / img_name+ img_path = test_dir / img_nameresult = engine(img_path, return_word_box=True)txts, _, _ = list(zip(*result.word_results[0]))assert txts == words@@ -47,14 +54,14 @@ def test_cn_word_ocr(engine, img_name: str, words: List[str]):[("issue_170.png", "TEST"), ("return_word_debug.jpg", "3F1")],)def test_en_word_ocr(engine, img_name: str, words: str):- img_path = tests_dir / img_name+ img_path = test_dir / img_nameresult = engine(img_path, return_word_box=True)txts, _, _ = list(zip(*result.word_results[0]))assert txts[0] == wordsdef test_en_return_single_char_box(engine):- img_path = tests_dir / "en.jpg"+ img_path = test_dir / "en.jpg"result = engine(img_path, return_word_box=True, return_single_char_box=True)txts, _, _ = list(zip(*result.word_results[0]))assert txts[0] == "3"