Skip to content

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

chore: update files

SWHLa86fb7381190
diff --git a/python/rapidocr/cal_rec_boxes/main.py b/python/rapidocr/cal_rec_boxes/main.py
index 43f53c2..8d6b9e3 100644
--- a/python/rapidocr/cal_rec_boxes/main.py
+++ b/python/rapidocr/cal_rec_boxes/main.py
@@ -12,7 +12,8 @@ from ..ch_ppocr_rec.utils import TextRecOutput
class CalRecBoxes:
- """计算识别文字的汉字单字和英文单词的坐标框。代码借鉴自PaddlePaddle/PaddleOCR和fanqie03/char-detection"""
+ """计算识别文字的汉字单字和英文单词的坐标框。
+ 代码借鉴自PaddlePaddle/PaddleOCR和fanqie03/char-detection"""
def __init__(self):
pass
@@ -43,7 +44,7 @@ class CalRecBoxes:
list(zip(word_box_content_list, conf_list, word_box_list))
)
- rec_res.word_results = tuple(list(v) for v in word_results)
+ rec_res.word_results = tuple(word_results)
return rec_res
@staticmethod
diff --git a/python/rapidocr/ch_ppocr_det/__init__.py b/python/rapidocr/ch_ppocr_det/__init__.py
index 502f4a3..7fa4067 100644
--- a/python/rapidocr/ch_ppocr_det/__init__.py
+++ b/python/rapidocr/ch_ppocr_det/__init__.py
@@ -1,4 +1,5 @@
# -*- encoding: utf-8 -*-
# @Author: SWHL
# @Contact: [email protected]
-from .text_detect import TextDetector
+from .main import TextDetector
+from .utils import TextDetOutput
diff --git a/python/rapidocr/ch_ppocr_det/text_detect.py b/python/rapidocr/ch_ppocr_det/main.py
similarity index 51%
rename from python/rapidocr/ch_ppocr_det/text_detect.py
rename to python/rapidocr/ch_ppocr_det/main.py
index c0c578e..a6cce33 100644
--- a/python/rapidocr/ch_ppocr_det/text_detect.py
+++ b/python/rapidocr/ch_ppocr_det/main.py
@@ -15,13 +15,13 @@
# @Author: SWHL
# @Contact: [email protected]
import time
-from typing import Any, Dict, Optional, Tuple
+from typing import Any, Dict, List
import numpy as np
from rapidocr.utils import OrtInferSession
-from .utils import DBPostProcess, DetPreProcess
+from .utils import DBPostProcess, DetPreProcess, TextDetOutput
class TextDetector:
@@ -44,7 +44,7 @@ class TextDetector:
self.infer = OrtInferSession(config)
- def __call__(self, img: np.ndarray) -> Tuple[Optional[np.ndarray], float]:
+ def __call__(self, img: np.ndarray) -> TextDetOutput:
start_time = time.perf_counter()
if img is None:
@@ -54,15 +54,18 @@ class TextDetector:
self.preprocess_op = self.get_preprocess(max(img.shape[0], img.shape[1]))
prepro_img = self.preprocess_op(img)
if prepro_img is None:
- return None, 0
+ return TextDetOutput()
preds = self.infer(prepro_img)[0]
- dt_boxes, dt_boxes_scores = self.postprocess_op(preds, ori_img_shape)
- dt_boxes = self.filter_tag_det_res(dt_boxes, ori_img_shape)
+ boxes, scores = self.postprocess_op(preds, ori_img_shape)
+ if len(boxes) < 1:
+ return TextDetOutput()
+
+ boxes = self.sorted_boxes(boxes)
elapse = time.perf_counter() - start_time
- return dt_boxes, elapse
+ return TextDetOutput(boxes, scores, elapse=elapse)
- def get_preprocess(self, max_wh):
+ def get_preprocess(self, max_wh: int) -> DetPreProcess:
if self.limit_type == "min":
limit_side_len = self.limit_side_len
elif max_wh < 960:
@@ -73,52 +76,28 @@ class TextDetector:
limit_side_len = 2000
return DetPreProcess(limit_side_len, self.limit_type, self.mean, self.std)
- def filter_tag_det_res(
- self, dt_boxes: np.ndarray, image_shape: Tuple[int, int]
- ) -> np.ndarray:
- img_height, img_width = image_shape
- dt_boxes_new = []
- for box in dt_boxes:
- box = self.order_points_clockwise(box)
- box = self.clip_det_res(box, img_height, img_width)
-
- rect_width = int(np.linalg.norm(box[0] - box[1]))
- rect_height = int(np.linalg.norm(box[0] - box[3]))
- if rect_width <= 3 or rect_height <= 3:
- continue
-
- dt_boxes_new.append(box)
- return np.array(dt_boxes_new)
-
- def order_points_clockwise(self, pts: np.ndarray) -> np.ndarray:
+ @staticmethod
+ def sorted_boxes(dt_boxes: np.ndarray) -> List[np.ndarray]:
"""
- reference from:
- https://github.com/jrosebr1/imutils/blob/master/imutils/perspective.py
- sort the points based on their x-coordinates
+ Sort text boxes in order from top to bottom, left to right
+ args:
+ dt_boxes(array):detected text boxes with shape [4, 2]
+ return:
+ sorted boxes(array) with shape [4, 2]
"""
- xSorted = pts[np.argsort(pts[:, 0]), :]
-
- # grab the left-most and right-most points from the sorted
- # x-roodinate points
- leftMost = xSorted[:2, :]
- rightMost = xSorted[2:, :]
-
- # now, sort the left-most coordinates according to their
- # y-coordinates so we can grab the top-left and bottom-left
- # points, respectively
- leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
- (tl, bl) = leftMost
-
- rightMost = rightMost[np.argsort(rightMost[:, 1]), :]
- (tr, br) = rightMost
-
- rect = np.array([tl, tr, br, bl], dtype="float32")
- return rect
-
- def clip_det_res(
- self, points: np.ndarray, img_height: int, img_width: int
- ) -> np.ndarray:
- for pno in range(points.shape[0]):
- points[pno, 0] = int(min(max(points[pno, 0], 0), img_width - 1))
- points[pno, 1] = int(min(max(points[pno, 1], 0), img_height - 1))
- return points
+ num_boxes = dt_boxes.shape[0]
+ sorted_boxes = sorted(dt_boxes, key=lambda x: (x[0][1], x[0][0]))
+ _boxes = list(sorted_boxes)
+
+ for i in range(num_boxes - 1):
+ for j in range(i, -1, -1):
+ if (
+ abs(_boxes[j + 1][0][1] - _boxes[j][0][1]) < 10
+ and _boxes[j + 1][0][0] < _boxes[j][0][0]
+ ):
+ tmp = _boxes[j]
+ _boxes[j] = _boxes[j + 1]
+ _boxes[j + 1] = tmp
+ else:
+ break
+ return _boxes
diff --git a/python/rapidocr/ch_ppocr_det/utils.py b/python/rapidocr/ch_ppocr_det/utils.py
index ef41063..377e9f2 100644
--- a/python/rapidocr/ch_ppocr_det/utils.py
+++ b/python/rapidocr/ch_ppocr_det/utils.py
@@ -1,6 +1,7 @@
# -*- encoding: utf-8 -*-
# @Author: SWHL
# @Contact: [email protected]
+from dataclasses import dataclass
from typing import List, Optional, Tuple
import cv2
@@ -9,6 +10,13 @@ import pyclipper
from shapely.geometry import Polygon
+@dataclass
+class TextDetOutput:
+ boxes: Optional[np.ndarray] = None
+ scores: Optional[Tuple[float]] = None
+ elapse: float = 0.0
+
+
class DetPreProcess:
def __init__(
self, limit_side_len: int = 736, limit_type: str = "min", mean=None, std=None
@@ -119,6 +127,7 @@ class DBPostProcess:
np.array(segmentation[0]).astype(np.uint8), self.dilation_kernel
)
boxes, scores = self.boxes_from_bitmap(pred[0], mask, src_w, src_h)
+ boxes, scores = self.filter_det_res(boxes, scores, src_h, src_w)
return boxes, scores
def boxes_from_bitmap(
@@ -235,3 +244,53 @@ class DBPostProcess:
offset.AddPath(box, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
expanded = np.array(offset.Execute(distance)).reshape((-1, 1, 2))
return expanded
+
+ def filter_det_res(
+ self, dt_boxes: np.ndarray, scores: List[float], img_height: int, img_width: int
+ ) -> Tuple[np.ndarray, List[float]]:
+ dt_boxes_new, new_scores = [], []
+ for box, score in zip(dt_boxes, scores):
+ box = self.order_points_clockwise(box)
+ box = self.clip_det_res(box, img_height, img_width)
+
+ rect_width = int(np.linalg.norm(box[0] - box[1]))
+ rect_height = int(np.linalg.norm(box[0] - box[3]))
+ if rect_width <= 3 or rect_height <= 3:
+ continue
+
+ dt_boxes_new.append(box)
+ new_scores.append(score)
+ return np.array(dt_boxes_new), new_scores
+
+ def order_points_clockwise(self, pts: np.ndarray) -> np.ndarray:
+ """
+ reference from:
+ https://github.com/jrosebr1/imutils/blob/master/imutils/perspective.py
+ sort the points based on their x-coordinates
+ """
+ xSorted = pts[np.argsort(pts[:, 0]), :]
+
+ # grab the left-most and right-most points from the sorted
+ # x-roodinate points
+ leftMost = xSorted[:2, :]
+ rightMost = xSorted[2:, :]
+
+ # now, sort the left-most coordinates according to their
+ # y-coordinates so we can grab the top-left and bottom-left
+ # points, respectively
+ leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
+ (tl, bl) = leftMost
+
+ rightMost = rightMost[np.argsort(rightMost[:, 1]), :]
+ (tr, br) = rightMost
+
+ rect = np.array([tl, tr, br, bl], dtype="float32")
+ return rect
+
+ def clip_det_res(
+ self, points: np.ndarray, img_height: int, img_width: int
+ ) -> np.ndarray:
+ for pno in range(points.shape[0]):
+ points[pno, 0] = int(min(max(points[pno, 0], 0), img_width - 1))
+ points[pno, 1] = int(min(max(points[pno, 1], 0), img_height - 1))
+ return points
diff --git a/python/rapidocr/ch_ppocr_rec/main.py b/python/rapidocr/ch_ppocr_rec/main.py
index 25d1bee..5926bc5 100644
--- a/python/rapidocr/ch_ppocr_rec/main.py
+++ b/python/rapidocr/ch_ppocr_rec/main.py
@@ -11,7 +11,6 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
-import argparse
import math
import time
from typing import Any, Dict
@@ -19,7 +18,7 @@ from typing import Any, Dict
import cv2
import numpy as np
-from rapidocr.utils import OrtInferSession, read_yaml
+from rapidocr.utils import OrtInferSession
from .utils import CTCLabelDecode, TextRecArguments, TextRecOutput
@@ -73,7 +72,7 @@ class TextRecognizer:
norm_img_batch.append(norm_img[np.newaxis, :])
norm_img_batch = np.concatenate(norm_img_batch).astype(np.float32)
- starttime = time.time()
+ start_time = time.perf_counter()
preds = self.session(norm_img_batch)[0]
line_results, word_results = self.postprocess_op(
preds,
@@ -88,7 +87,7 @@ class TextRecognizer:
continue
rec_res[indices[beg_img_no + rno]] = (one_res, None)
- elapse += time.time() - starttime
+ elapse += time.perf_counter() - start_time
all_line_results, all_word_results = list(zip(*rec_res))
return TextRecOutput(all_line_results, all_word_results, elapse)
@@ -115,17 +114,3 @@ class TextRecognizer:
padding_im = np.zeros((img_channel, img_height, img_width), dtype=np.float32)
padding_im[:, :, 0:resized_w] = resized_image
return padding_im
-
-
-if __name__ == "__main__":
- parser = argparse.ArgumentParser()
- parser.add_argument("--image_path", type=str, help="image_dir|image_path")
- parser.add_argument("--config_path", type=str, default="config.yaml")
- args = parser.parse_args()
-
- config = read_yaml(args.config_path)
- text_recognizer = TextRecognizer(config)
-
- img = cv2.imread(args.image_path)
- rec_res, predict_time = text_recognizer(img)
- print(f"rec result: {rec_res}\t cost: {predict_time}s")
diff --git a/python/rapidocr/ch_ppocr_rec/utils.py b/python/rapidocr/ch_ppocr_rec/utils.py
index 7644201..6130f59 100644
--- a/python/rapidocr/ch_ppocr_rec/utils.py
+++ b/python/rapidocr/ch_ppocr_rec/utils.py
@@ -29,8 +29,10 @@ class TextRecArguments:
@dataclass
class TextRecOutput:
- line_results: Tuple[List[Any]] = ([""],)
- word_results: Tuple[List[Any]] = ([""],)
+ line_results: Tuple[Tuple[str, float]] = (("", 1.0),)
+ word_results: Tuple[Tuple[str, float, Optional[List[List[int]]]]] = (
+ ("", 1.0, None),
+ )
elapse: Optional[float] = None
diff --git a/python/rapidocr/main.py b/python/rapidocr/main.py
index 0172735..efe9c74 100644
--- a/python/rapidocr/main.py
+++ b/python/rapidocr/main.py
@@ -14,6 +14,7 @@ from .ch_ppocr_det import TextDetector
from .ch_ppocr_rec import TextRecArguments, TextRecognizer, TextRecOutput
from .utils import (
LoadImage,
+ RapidOCROutput,
UpdateParameters,
VisRes,
add_round_letterbox,
@@ -93,14 +94,14 @@ class RapidOCR:
img, ratio_h, ratio_w = self.preprocess(img)
op_record["preprocess"] = {"ratio_h": ratio_h, "ratio_w": ratio_w}
- dt_boxes, cls_res, rec_res = None, None, None
- det_elapse, cls_elapse, rec_elapse = 0.0, 0.0, 0.0
+ dt_boxes, cls_res, rec_res = None, None, TextRecOutput()
+ det_elapse, cls_elapse = 0.0, 0.0
if use_det:
img, op_record = self.maybe_add_letterbox(img, op_record)
- dt_boxes, det_elapse = self.auto_text_det(img)
- if dt_boxes is None:
- return None, None
+ det_res = self.text_det(img)
+ if det_res.boxes is None:
+ return RapidOCROutput()
img = self.get_crop_img_list(img, dt_boxes)
@@ -117,13 +118,18 @@ class RapidOCR:
and all(v for v in rec_res.word_results)
):
rec_res = self.cal_rec_boxes(img, dt_boxes, rec_res)
- for rec_res_i in rec_res.word_results:
- if rec_res_i[2]:
- rec_res_i[2] = (
- self._get_origin_points([rec_res_i[2]], op_record, raw_h, raw_w)
- .astype(np.int32)
- .tolist()
- )
+ origin_words = []
+ for one_word in rec_res.word_results:
+ one_word_points = one_word[2]
+ if one_word_points is None:
+ continue
+
+ origin_words_points = self._get_origin_points(
+ [one_word_points], op_record, raw_h, raw_w
+ )
+ origin_words_points = origin_words_points.astype(np.int32).tolist()[0]
+ origin_words.append((one_word[0], one_word[1], origin_words_points))
+ rec_res.word_results = tuple(origin_words)
if dt_boxes is not None:
dt_boxes = self._get_origin_points(dt_boxes, op_record, raw_h, raw_w)
diff --git a/python/rapidocr/utils/__init__.py b/python/rapidocr/utils/__init__.py
index 150da63..0dde27c 100644
--- a/python/rapidocr/utils/__init__.py
+++ b/python/rapidocr/utils/__init__.py
@@ -11,6 +11,7 @@ from .load_image import LoadImage, LoadImageError
from .logger import get_logger
from .parse_parameters import UpdateParameters, init_args, update_model_path
from .process_img import add_round_letterbox, increase_min_side, reduce_max_side
+from .typings import RapidOCROutput
from .vis_res import VisRes
diff --git a/python/rapidocr/utils/typings.py b/python/rapidocr/utils/typings.py
new file mode 100644
index 0000000..b3e2907
--- /dev/null
+++ b/python/rapidocr/utils/typings.py
@@ -0,0 +1,9 @@
+# -*- encoding: utf-8 -*-
+# @Author: SWHL
+# @Contact: [email protected]
+from dataclasses import dataclass
+
+
+@dataclass
+class RapidOCROutput:
+ pass