Skip to content

📄 Awesome OCR multiple programing languages toolkits based on ONNX Runtime, OpenVINO, MNN, PaddlePaddle, TensorRT and PyTorch.

fix: correct incorrect character coordinates in returned data (#614)

SWHLdf8c571f1cd8
diff --git a/python/rapidocr/main.py b/python/rapidocr/main.py
index ceadbbc..a2890dd 100644
--- a/python/rapidocr/main.py
+++ b/python/rapidocr/main.py
@@ -23,9 +23,11 @@ from .utils.process_img import (
apply_vertical_padding,
get_rotate_crop_image,
map_boxes_to_original,
+ map_img_to_original,
resize_image_within_bounds,
)
from .utils.typings import LangRec
+from .utils.utils import filter_by_indices
from .utils.vis_res import VisRes
root_dir = Path(__file__).resolve().parent
@@ -159,28 +161,27 @@ class RapidOCR:
det_res.boxes, op_record, ori_h, ori_w
)
+ ratio_h = op_record["preprocess"]["ratio_h"]
+ ratio_w = op_record["preprocess"]["ratio_w"]
+ cropped_img_list = map_img_to_original(cropped_img_list, ratio_h, ratio_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()}
+ valid_ids = [i for i, v in enumerate(rec_res.txts) if 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
- ]
+ det_res.boxes = filter_by_indices(det_res.boxes, valid_ids)
+ det_res.scores = filter_by_indices(det_res.scores, valid_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
- ]
- rec_res.scores = [
- v for i, v in enumerate(rec_res.scores) if i not in empty_ids
- ]
+ rec_res.imgs = filter_by_indices(rec_res.imgs, valid_ids)
+ rec_res.txts = filter_by_indices(rec_res.txts, valid_ids)
+ rec_res.word_results = filter_by_indices(rec_res.word_results, valid_ids)
+ rec_res.scores = filter_by_indices(rec_res.scores, valid_ids)
+
+ cropped_img_list = filter_by_indices(cropped_img_list, valid_ids)
# 仅分类结果
if (
@@ -243,17 +244,9 @@ class RapidOCR:
origin_words = []
for word_line in rec_res.word_results:
- origin_words_item = []
- for txt, score, bbox in word_line:
- if bbox is None:
- continue
-
- 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))
-
+ origin_words_item = [
+ (txt, score, bbox) for txt, score, bbox in word_line if bbox is not None
+ ]
if origin_words_item:
origin_words.append(tuple(origin_words_item))
return tuple(origin_words)
diff --git a/python/rapidocr/utils/process_img.py b/python/rapidocr/utils/process_img.py
index 46e8a16..13a2804 100644
--- a/python/rapidocr/utils/process_img.py
+++ b/python/rapidocr/utils/process_img.py
@@ -1,12 +1,24 @@
# -*- encoding: utf-8 -*-
# @Author: SWHL
# @Contact: [email protected]
-from typing import Any, Dict, Tuple
+from typing import Any, Dict, List, Tuple
import cv2
import numpy as np
+def map_img_to_original(
+ imgs: List[np.ndarray], ratio_h: float, ratio_w: float
+) -> List[np.ndarray]:
+ results = []
+ for img in imgs:
+ img_h, img_w = img.shape[:2]
+ ori_img_h, ori_img_w = round(img_h * ratio_h), round(img_w * ratio_w)
+ resize_img = cv2.resize(img, (ori_img_w, ori_img_h))
+ results.append(resize_img)
+ return results
+
+
def map_boxes_to_original(
dt_boxes: np.ndarray, op_record: Dict[str, Any], ori_h: int, ori_w: int
) -> np.ndarray:
diff --git a/python/rapidocr/utils/utils.py b/python/rapidocr/utils/utils.py
index c0f63a3..54ea641 100644
--- a/python/rapidocr/utils/utils.py
+++ b/python/rapidocr/utils/utils.py
@@ -5,13 +5,26 @@ import hashlib
import importlib
from pathlib import Path
from sys import platform
-from typing import Tuple, Union
+from typing import Any, List, Tuple, Union
from urllib.parse import urlparse
import cv2
import numpy as np
+def filter_by_indices(
+ data: Union[np.ndarray, List[Any], Tuple[Any]],
+ indices: Union[np.ndarray, List[int], Tuple[int, ...]],
+) -> Union[np.ndarray, List[Any], Tuple[Any]]:
+ if isinstance(data, np.ndarray):
+ return data[indices]
+
+ if isinstance(data, (list, tuple)):
+ return [data[i] for i in indices]
+
+ raise TypeError(f"Unsupported data type: {type(data)}")
+
+
def mkdir(dir_path):
Path(dir_path).mkdir(parents=True, exist_ok=True)
@@ -29,8 +42,16 @@ def quads_to_rect_bbox(bbox: np.ndarray) -> Tuple[float, float, float, float]:
return float(x_min), float(y_min), float(x_max), float(y_max)
+def is_chinese_char(ch: str) -> bool:
+ return (
+ "\u4e00" <= ch <= "\u9fff" # 汉字
+ or "\u3000" <= ch <= "\u303f" # CJK 标点(如 。 、 “” 《》 ……)
+ or "\uff00" <= ch <= "\uffef" # 全角符号(如 ,.!?【】)
+ )
+
+
def has_chinese_char(text: str) -> bool:
- return any("\u4e00" <= ch <= "\u9fff" for ch in text)
+ return any(is_chinese_char(ch) for ch in text)
def get_file_sha256(file_path: Union[str, Path], chunk_size: int = 65536) -> str: