Skip to content

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

feat(rapidocr): support to other language ocr and update visualization (#369)

SWHL202dfb91fc43
diff --git a/python/demo.py b/python/demo.py
index aec145d..295c740 100644
--- a/python/demo.py
+++ b/python/demo.py
@@ -10,5 +10,3 @@ result = engine(img_url)
print(result)
result.vis()
-
-engine.export_config("1.yaml")
diff --git a/python/rapidocr/ch_ppocr_cls/utils.py b/python/rapidocr/ch_ppocr_cls/utils.py
index f2e09df..7b9c229 100644
--- a/python/rapidocr/ch_ppocr_cls/utils.py
+++ b/python/rapidocr/ch_ppocr_cls/utils.py
@@ -12,10 +12,17 @@
# See the License for the specific language governing permissions and
# limitations under the License.
from dataclasses import dataclass
-from typing import List, Optional, Tuple
+from pathlib import Path
+from typing import List, Optional, Tuple, Union
import numpy as np
+from ..utils.logger import Logger
+from ..utils.utils import save_img
+from ..utils.vis_res import VisRes
+
+logger = Logger(logger_name=__name__).get_log()
+
@dataclass
class TextClsOutput:
@@ -28,6 +35,22 @@ class TextClsOutput:
return 0
return len(self.img_list)
+ def vis(self, save_path: Optional[Union[str, Path]] = None) -> Optional[np.ndarray]:
+ if self.img_list is None or self.cls_res is None:
+ logger.warning("No image or txts to visualize.")
+ return None
+
+ txts = [f"{txt} {score:.2f}" for txt, score in self.cls_res]
+ scores = [score for _, score in self.cls_res]
+
+ vis = VisRes()
+ vis_img = vis.draw_rec_res(self.img_list, txts, scores)
+
+ if save_path is not None:
+ save_img(save_path, vis_img)
+ logger.info("Visualization saved as %s", save_path)
+ return vis_img
+
class ClsPostProcess:
def __init__(self, label_list: List[str]):
diff --git a/python/rapidocr/ch_ppocr_det/main.py b/python/rapidocr/ch_ppocr_det/main.py
index c432eb0..cb7e970 100644
--- a/python/rapidocr/ch_ppocr_det/main.py
+++ b/python/rapidocr/ch_ppocr_det/main.py
@@ -63,7 +63,7 @@ class TextDetector:
boxes = self.sorted_boxes(boxes)
elapse = time.perf_counter() - start_time
- return TextDetOutput(boxes, scores, elapse=elapse)
+ return TextDetOutput(img, boxes, scores, elapse=elapse)
def get_preprocess(self, max_wh: int) -> DetPreProcess:
if self.limit_type == "min":
diff --git a/python/rapidocr/ch_ppocr_det/utils.py b/python/rapidocr/ch_ppocr_det/utils.py
index 8210428..4690b06 100644
--- a/python/rapidocr/ch_ppocr_det/utils.py
+++ b/python/rapidocr/ch_ppocr_det/utils.py
@@ -9,9 +9,16 @@ import numpy as np
import pyclipper
from shapely.geometry import Polygon
+from ..utils.logger import Logger
+from ..utils.utils import save_img
+from ..utils.vis_res import VisRes
+
+logger = Logger(logger_name=__name__).get_log()
+
@dataclass
class TextDetOutput:
+ img: Optional[np.ndarray] = None
boxes: Optional[np.ndarray] = None
scores: Optional[Tuple[float]] = None
elapse: float = 0.0
@@ -21,6 +28,19 @@ class TextDetOutput:
return 0
return len(self.boxes)
+ def vis(self, save_path: Optional[str] = None) -> Optional[np.ndarray]:
+ if self.img is None or self.boxes is None or self.scores is None:
+ logger.warning("No image or boxes to visualize.")
+ return None
+
+ vis = VisRes()
+ vis_img = vis.draw_dt_boxes(self.img, self.boxes, self.scores)
+
+ if save_path is not None:
+ save_img(save_path, vis_img)
+ logger.info("Visualization saved as %s", save_path)
+ return vis_img
+
class DetPreProcess:
def __init__(
diff --git a/python/rapidocr/ch_ppocr_rec/main.py b/python/rapidocr/ch_ppocr_rec/main.py
index 29a4076..7d7f988 100644
--- a/python/rapidocr/ch_ppocr_rec/main.py
+++ b/python/rapidocr/ch_ppocr_rec/main.py
@@ -101,7 +101,7 @@ class TextRecognizer:
all_line_results, all_word_results = list(zip(*rec_res))
txts, scores = list(zip(*all_line_results))
- return TextRecOutput(txts, scores, all_word_results, elapse)
+ return TextRecOutput(img_list, txts, scores, all_word_results, elapse)
def resize_norm_img(self, img: np.ndarray, max_wh_ratio: float) -> np.ndarray:
img_channel, img_height, img_width = self.rec_image_shape
diff --git a/python/rapidocr/ch_ppocr_rec/utils.py b/python/rapidocr/ch_ppocr_rec/utils.py
index f06c71e..4b4ec15 100644
--- a/python/rapidocr/ch_ppocr_rec/utils.py
+++ b/python/rapidocr/ch_ppocr_rec/utils.py
@@ -7,6 +7,12 @@ from typing import Any, List, Optional, Tuple, Union
import numpy as np
+from ..utils.logger import Logger
+from ..utils.utils import save_img
+from ..utils.vis_res import VisRes
+
+logger = Logger(logger_name=__name__).get_log()
+
@dataclass
class TextRecConfig:
@@ -29,18 +35,35 @@ class TextRecInput:
@dataclass
class TextRecOutput:
+ imgs: Optional[List[np.ndarray]] = None
txts: Optional[Tuple[str]] = None
scores: Tuple[float] = (1.0,)
word_results: Tuple[Tuple[str, float, Optional[List[List[int]]]]] = (
("", 1.0, None),
)
elapse: Optional[float] = None
+ lang_rec: Optional[str] = None
def __len__(self):
if self.txts is None:
return 0
return len(self.txts)
+ def vis(self, save_path: Optional[Union[str, Path]] = None) -> Optional[np.ndarray]:
+ if self.imgs is None or self.txts is None:
+ logger.warning("No image or txts to visualize.")
+ return None
+
+ vis = VisRes()
+ vis_img = vis.draw_rec_res(
+ self.imgs, self.txts, self.scores, lang_rec=self.lang_rec
+ )
+
+ if save_path is not None:
+ save_img(save_path, vis_img)
+ logger.info("Visualization saved as %s", save_path)
+ return vis_img
+
class CTCLabelDecode:
def __init__(
diff --git a/python/rapidocr/default_models.yaml b/python/rapidocr/default_models.yaml
index 9a687bd..c9c0988 100644
--- a/python/rapidocr/default_models.yaml
+++ b/python/rapidocr/default_models.yaml
@@ -4,7 +4,7 @@ onnxruntime:
ch_PP-OCRv4_det_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/ch_PP-OCRv4_det_infer.onnx
ch_PP-OCRv4_det_server_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/ch_PP-OCRv4_det_server_infer.onnx
en_PP-OCRv3_det_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/en_PP-OCRv3_det_infer.onnx
- Multilingual_PP-OCRv3_det_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/Multilingual_PP-OCRv3_det_infer.onnx
+ multi_PP-OCRv3_det_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/Multilingual_PP-OCRv3_det_infer.onnx
rec:
arabic_PP-OCRv4_rec_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/rec/arabic_PP-OCRv4_rec_infer.onnx
ch_PP-OCRv4_rec_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/rec/ch_PP-OCRv4_rec_infer.onnx
@@ -28,7 +28,7 @@ openvino:
ch_PP-OCRv4_det_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/ch_PP-OCRv4_det_infer.onnx
ch_PP-OCRv4_det_server_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/ch_PP-OCRv4_det_server_infer.onnx
en_PP-OCRv3_det_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/en_PP-OCRv3_det_infer.onnx
- Multilingual_PP-OCRv3_det_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/Multilingual_PP-OCRv3_det_infer.onnx
+ multi_PP-OCRv3_det_infer.onnx: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/det/Multilingual_PP-OCRv3_det_infer.onnx
rec:
arabic_PP-OCRv4_rec_infer:
model_dir: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/onnx/PP-OCRv4/rec/arabic_PP-OCRv4_rec_infer.onnx
@@ -78,7 +78,7 @@ paddlepaddle:
ch_PP-OCRv4_det_infer: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/paddle/PP-OCRv4/det/ch_PP-OCRv4_det_infer
ch_PP-OCRv4_det_server_infer: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/paddle/PP-OCRv4/det/ch_PP-OCRv4_det_server_infer
en_PP-OCRv3_det_infer: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/paddle/PP-OCRv4/det/en_PP-OCRv3_det_infer
- Multilingual_PP-OCRv3_det_infer: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/paddle/PP-OCRv4/det/Multilingual_PP-OCRv3_det_infer
+ multi_PP-OCRv3_det_infer: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/paddle/PP-OCRv4/det/Multilingual_PP-OCRv3_det_infer
rec:
arabic_PP-OCRv4_rec_infer:
model_dir: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/paddle/PP-OCRv4/rec/arabic_PP-OCRv4_rec_infer
@@ -136,3 +136,17 @@ torch:
dict_url: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/paddle/PP-OCRv4/rec/ch_PP-OCRv4_rec_infer/ppocr_keys_v1.txt
cls:
ch_ptocr_mobile_v2.0_cls_infer.pth: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/torch/ch_ptocr_mobile_v2.0_cls_infer.pth
+
+fonts:
+ ch: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/FZYTK.TTF
+ en: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/FZYTK.TTF
+ japan: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/japan.ttc
+ arabic: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/arabic.ttf
+ chinese_cht: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/chinese_cht.ttf
+ cyrillic: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/cyrillic.ttf
+ devanagari: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/devanagari_Martel-Regular.ttf
+ korean: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/korean.ttf
+ ka: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/kannada.ttf
+ latin: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/latin.ttf
+ ta: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/tamil.ttf
+ te: https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/telugu.ttf
\ No newline at end of file
diff --git a/python/rapidocr/main.py b/python/rapidocr/main.py
index 2c4b264..b57ff20 100644
--- a/python/rapidocr/main.py
+++ b/python/rapidocr/main.py
@@ -46,6 +46,8 @@ class RapidOCR:
lang_det, lang_rec = ParseLang()(config.Global.lang_det, config.Global.lang_rec)
+ self.lang_rec = lang_rec
+
self.text_score = config.Global.text_score
self.min_height = config.Global.min_height
self.width_height_ratio = config.Global.width_height_ratio
@@ -92,7 +94,7 @@ class RapidOCR:
text_score: float = 0.5,
box_thresh: float = 0.5,
unclip_ratio: float = 1.6,
- ) -> RapidOCROutput:
+ ) -> Union[TextDetOutput, TextClsOutput, TextRecOutput, RapidOCROutput]:
use_det = self.use_det if use_det is None else use_det
use_cls = self.use_cls if use_cls is None else use_cls
use_rec = self.use_rec if use_rec is None else use_rec
@@ -151,6 +153,7 @@ class RapidOCR:
det_res.boxes, op_record, raw_h, raw_w
)
+ rec_res.lang_rec = self.lang_rec
ocr_res = self.get_final_res(ori_img, det_res, cls_res, rec_res)
return ocr_res
@@ -293,6 +296,7 @@ class RapidOCR:
scores=rec_res.scores,
word_results=rec_res.word_results,
elapse_list=[det_res.elapse, cls_res.elapse, rec_res.elapse],
+ lang_rec=self.lang_rec,
)
ocr_res = self.filter_by_text_score(ocr_res)
if len(ocr_res) <= 0:
diff --git a/python/rapidocr/utils/parse_parameters.py b/python/rapidocr/utils/parse_parameters.py
index d1ec299..f3a6d3a 100644
--- a/python/rapidocr/utils/parse_parameters.py
+++ b/python/rapidocr/utils/parse_parameters.py
@@ -14,117 +14,49 @@ InputType = Union[str, np.ndarray, bytes, Path, Image.Image]
class ParseLang:
def __init__(self):
- self.latin_lang = [
- "af",
- "az",
- "bs",
- "cs",
- "cy",
- "da",
- "de",
- "es",
- "et",
- "fr",
- "ga",
- "hr",
- "hu",
- "id",
- "is",
- "it",
- "ku",
- "la",
- "lt",
- "lv",
- "mi",
- "ms",
- "mt",
- "nl",
- "no",
- "oc",
- "pi",
- "pl",
- "pt",
- "ro",
- "rs_latin",
- "sk",
- "sl",
- "sq",
- "sv",
- "sw",
- "tl",
- "tr",
- "uz",
- "vi",
- "french",
- "german",
- ]
-
- self.arabic_lang = ["ar", "fa", "ug", "ur"]
- self.cyrillic_lang = [
- "ru",
- "rs_cyrillic",
- "be",
- "bg",
- "uk",
- "mn",
- "abq",
- "ady",
- "kbd",
- "ava",
- "dar",
- "inh",
- "che",
- "lbe",
- "lez",
- "tab",
- ]
- self.devanagari_lang = [
- "hi",
- "mr",
- "ne",
- "bh",
- "mai",
- "ang",
- "bho",
- "mah",
- "sck",
- "new",
- "gom",
- "sa",
- "bgc",
+ self.lang_det_list = ["ch", "en", "multi"]
+ self.lang_rec_list = [
+ "ch",
+ "en",
+ "arabic",
+ "chinese_cht",
+ "cyrillic",
+ "devanagari",
+ "japan",
+ "korean",
+ "ka",
+ "latin",
+ "ta",
+ "te",
]
def __call__(self, lang_det: str, lang_rec: str) -> Tuple[str, str]:
- lang_det, det_model_type = lang_det.split("_")
- lang_rec, rec_model_type = lang_rec.split("_")
- return f"{lang_det}_{det_model_type}", f"{lang_rec}_{rec_model_type}"
+ lang_det, det_model_type = lang_det.rsplit("_", 1)
+ lang_det = self.parse_det_lang(lang_det)
- def parse_det_lang(self, lang: str) -> str:
- if lang == "ch":
- return "ch"
+ lang_rec, rec_model_type = lang_rec.rsplit("_", 1)
+ lang_rec = self.parse_rec_lang(lang_rec)
- if lang in ["en", "latin"]:
- return "en"
-
- return "Multilingual"
-
- def parse_rec_lang(self, lang: str):
- if lang in self.latin_lang:
- return "latin"
+ return f"{lang_det}_{det_model_type}", f"{lang_rec}_{rec_model_type}"
- if lang in self.arabic_lang:
- return "arabic"
+ def parse_det_lang(self, lang: str) -> str:
+ lang = lang.strip().lower()
- if lang in self.cyrillic_lang:
- return "cyrillic"
+ if lang in self.lang_det_list:
+ return lang
+ raise ValueError(
+ f"lang: {lang} is not in the supported list: {self.lang_det_list}"
+ )
- if lang in self.devanagari_lang:
- return "devanagari"
+ def parse_rec_lang(self, lang: str) -> str:
+ lang = lang.strip().lower()
- if lang in ["ch", "en"]:
- pass
+ if lang in self.lang_rec_list:
+ return lang
- raise ValueError(f"lang: {lang} is not in the supported list.")
+ raise ValueError(
+ f"lang: {lang} is not in the supported list: {self.lang_rec_list}"
+ )
class ParseParams(OmegaConf):
diff --git a/python/rapidocr/utils/typings.py b/python/rapidocr/utils/typings.py
index a843ea3..55e0fba 100644
--- a/python/rapidocr/utils/typings.py
+++ b/python/rapidocr/utils/typings.py
@@ -4,10 +4,10 @@
from dataclasses import dataclass, field
from typing import List, Optional, Tuple, Union
-import cv2
import numpy as np
from .logger import Logger
+from .utils import save_img
from .vis_res import VisRes
logger = Logger(logger_name=__name__).get_log()
@@ -24,6 +24,7 @@ class RapidOCROutput:
)
elapse_list: List[Union[float, None]] = field(default_factory=list)
elapse: float = field(init=False)
+ lang_rec: Optional[str] = None
def __post_init__(self):
self.elapse = sum(v for v in self.elapse_list if isinstance(v, float))
@@ -52,22 +53,41 @@ class RapidOCROutput:
final_res.append([box, list(rec)])
return final_res
- def vis(self):
+ def vis(self, save_path: Optional[str] = None, font_path: Optional[str] = None):
if self.img is None or self.boxes is None:
logger.warning("No image or boxes to visualize.")
return
vis = VisRes()
if all(v is None for v in self.word_results):
- vis_img = vis(self.img, self.boxes, self.txts, self.scores)
- cv2.imwrite("vis.png", vis_img)
- logger.info("Visualization saved as vis.png.")
+ vis_img = vis(
+ self.img,
+ self.boxes,
+ self.txts,
+ self.scores,
+ font_path=font_path,
+ lang_rec=self.lang_rec,
+ )
+ if save_path is None:
+ save_path = "vis_det_cls_rec_result.png"
+
+ save_img(save_path, vis_img)
+ logger.info("Visualization saved as %s", save_path)
return
# single word vis
words_results = self.word_results
words, words_scores, words_boxes = list(zip(*words_results))
- vis_img = vis(self.img, words_boxes, words, words_scores)
- cv2.imwrite("vis_single.png", vis_img)
- logger.info("Single word visualization saved as vis_single.png.")
- return
+ vis_img = vis(
+ self.img,
+ words_boxes,
+ words,
+ words_scores,
+ font_path=font_path,
+ lang_rec=self.lang_rec,
+ )
+ if save_path is None:
+ save_single_path = "vis_single.png"
+
+ save_img(save_single_path, vis_img)
+ logger.info("Single word visualization saved as %s", save_single_path)
diff --git a/python/rapidocr/utils/utils.py b/python/rapidocr/utils/utils.py
index e4e0fee..99416cc 100644
--- a/python/rapidocr/utils/utils.py
+++ b/python/rapidocr/utils/utils.py
@@ -6,10 +6,19 @@ from pathlib import Path
from typing import Union
from urllib.parse import urlparse
+import cv2
+import numpy as np
import requests
from tqdm import tqdm
+def save_img(save_path: Union[str, Path], img: np.ndarray):
+ if not Path(save_path).parent.exists():
+ Path(save_path).parent.mkdir(parents=True, exist_ok=True)
+
+ cv2.imwrite(str(save_path), img)
+
+
def is_url(url: str) -> bool:
try:
result = urlparse(url)
diff --git a/python/rapidocr/utils/vis_res.py b/python/rapidocr/utils/vis_res.py
index 94598dd..eba6fe4 100644
--- a/python/rapidocr/utils/vis_res.py
+++ b/python/rapidocr/utils/vis_res.py
@@ -8,23 +8,30 @@ from typing import List, Optional, Tuple, Union
import cv2
import numpy as np
+from omegaconf import OmegaConf
from PIL import Image, ImageDraw, ImageFont
from .load_image import LoadImage
+from .logger import Logger
from .utils import download_file
root_dir = Path(__file__).resolve().parent.parent
InputType = Union[str, np.ndarray, bytes, Path, Image.Image]
-DEFAULT_FONT_PATH = root_dir / "models" / "FZYTK.TTF"
-DEFAULT_FONT_URL = "https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/fonts/FZYTK.TTF"
+DEFAULT_FONT_DIR = root_dir / "models"
+DEFAULT_FONT_PATH = DEFAULT_FONT_DIR / "FZYTK.TTF"
+FONT_YAML_PATH = root_dir / "default_models.yaml"
class VisRes:
def __init__(self, text_score: float = 0.5):
+ self.logger = Logger(logger_name=__name__).get_log()
+
self.text_score = text_score
self.load_img = LoadImage()
+ self.font_cfg = OmegaConf.load(FONT_YAML_PATH).fonts
+
def __call__(
self,
img_content: InputType,
@@ -32,15 +39,25 @@ class VisRes:
txts: Optional[Union[List[str], Tuple[str]]] = None,
scores: Optional[Tuple[float]] = None,
font_path: Optional[str] = None,
+ lang_rec: Optional[str] = None,
) -> np.ndarray:
if txts is None:
- return self.draw_dt_boxes(img_content, dt_boxes)
- return self.draw_ocr_box_txt(img_content, dt_boxes, txts, scores, font_path)
+ return self.draw_dt_boxes(img_content, dt_boxes, scores)
- def draw_dt_boxes(self, img_content: InputType, dt_boxes: np.ndarray) -> np.ndarray:
+ font_path = self.get_font_path(font_path, lang_rec)
+ return self.draw_ocr_box_txt(img_content, dt_boxes, txts, font_path, scores)
+
+ def draw_dt_boxes(
+ self,
+ img_content: InputType,
+ dt_boxes: np.ndarray,
+ scores: Optional[Tuple[float]] = None,
+ ) -> np.ndarray:
img = self.load_img(img_content)
+ if scores is None:
+ scores = [1.0] * len(dt_boxes)
- for idx, box in enumerate(dt_boxes):
+ for idx, (box, score) in enumerate(zip(dt_boxes, scores)):
color = self.get_random_color()
points = np.array(box)
@@ -48,20 +65,107 @@ class VisRes:
start_point = round(points[0][0]), round(points[0][1])
cv2.putText(
- img, f"{idx}", start_point, cv2.FONT_HERSHEY_SIMPLEX, 1, color, 3
+ img,
+ f"{idx}:{score:.3f}",
+ start_point,
+ cv2.FONT_HERSHEY_SIMPLEX,
+ 1,
+ color,
+ 3,
)
return img
+ def get_font_path(
+ self,
+ font_path: Optional[Union[str, Path]] = None,
+ lang_rec: Optional[str] = None,
+ ) -> str:
+ if lang_rec is None:
+ # 没有指定语种,用默认字体文件
+ download_file(self.font_cfg["ch"], DEFAULT_FONT_PATH, logger=self.logger)
+ return str(DEFAULT_FONT_PATH)
+
+ if font_path is None:
+ # 指定了语种,但是没有指定字体文件,根据语种选择字体文件
+ lang_rec = lang_rec.rsplit("_", 1)[0]
+ font_url = self.font_cfg.get(lang_rec, None)
+ if font_url is None:
+ self.logger.warning(
+ "Font file for %s is not found in the supported font list. Default font file will be used.",
+ lang_rec,
+ )
+ download_file(
+ self.font_cfg["ch"], DEFAULT_FONT_PATH, logger=self.logger
+ )
+ return str(DEFAULT_FONT_PATH)
+
+ save_font_path = DEFAULT_FONT_DIR / f"{Path(font_url).name}"
+ download_file(font_url, save_font_path, logger=self.logger)
+ return str(save_font_path)
+
+ return str(font_path)
+
+ def draw_rec_res(
+ self,
+ imgs: List[InputType],
+ txts: Union[List[str], Tuple[str]],
+ scores: Tuple[float],
+ lang_rec: Optional[str] = None,
+ ) -> np.ndarray:
+ result_imgs = []
+ for img, txt, score in zip(imgs, txts, scores):
+ vis_img = self.draw_one_rec_res(img, txt, score, lang_rec)
+ result_imgs.append(vis_img)
+ return self.concat_imgs(result_imgs, direction="vertical")
+
+ def draw_one_rec_res(
+ self,
+ img_content: InputType,
+ txt: str,
+ score: float,
+ lang_rec: Optional[str] = None,
+ ) -> np.ndarray:
+ font_path = self.get_font_path(None, lang_rec)
+
+ image = Image.fromarray(self.load_img(img_content))
+ h, w = image.height, image.width
+ if image.mode == "L":
+ image = image.convert("RGB")
+
+ img_left = image.copy()
+ img_right = Image.new("RGB", (w, h), (255, 255, 255))
+ draw_right = ImageDraw.Draw(img_right)
+ box = [[0, 0], [w, 0], [w, h], [0, h]]
+
+ box_height = self.get_box_height(box)
+ box_width = self.get_box_width(box)
+ if box_height > 2 * box_width:
+ font_size = max(int(box_width * 0.9), 10)
+ font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
+ cur_y = box[0][1]
+
+ for c in txt:
+ draw_right.text((box[0][0] + 3, cur_y), c, fill=(0, 0, 0), font=font)
+ cur_y += self.get_char_size(font, c)
+ else:
+ font_size = max(int(box_height * 0.8), 10)
+ font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
+ draw_right.text([box[0][0], box[0][1]], txt, fill=(0, 0, 0), font=font)
+
+ img_left = Image.blend(image, img_left, 0.5)
+ img_show = Image.new("RGB", (w * 2, h), 255)
+ img_show.paste(img_left, (0, 0, w, h))
+ img_show.paste(img_right, (w, 0, w * 2, h))
+ return np.array(img_show)
+
def draw_ocr_box_txt(
self,
img_content: InputType,
dt_boxes: np.ndarray,
txts: Union[List[str], Tuple[str]],
+ font_path: str,
scores: Optional[Tuple[float]] = None,
- font_path: Optional[str] = None,
) -> np.ndarray:
- font_path = self.get_font_path(font_path)
-
image = Image.fromarray(self.load_img(img_content))
h, w = image.height, image.width
if image.mode == "L":
@@ -106,13 +210,6 @@ class VisRes:
img_show.paste(img_right, (w, 0, w * 2, h))
return np.array(img_show)
- @staticmethod
- def get_font_path(font_path: Optional[Union[str, Path]] = None) -> str:
- if font_path is None or not Path(font_path).exists():
- download_file(DEFAULT_FONT_URL, DEFAULT_FONT_PATH, logger=None)
- return str(DEFAULT_FONT_PATH)
- return str(font_path)
-
@staticmethod
def get_random_color() -> Tuple[int, int, int]:
return (
@@ -122,11 +219,11 @@ class VisRes:
)
@staticmethod
- def get_box_height(box: List[List[float]]) -> float:
+ def get_box_height(box: List[List[Union[float, int]]]) -> float:
return math.sqrt((box[0][0] - box[3][0]) ** 2 + (box[0][1] - box[3][1]) ** 2)
@staticmethod
- def get_box_width(box: List[List[float]]) -> float:
+ def get_box_width(box: List[List[Union[float, int]]]) -> float:
return math.sqrt((box[0][0] - box[1][0]) ** 2 + (box[0][1] - box[1][1]) ** 2)
@staticmethod
@@ -143,3 +240,36 @@ class VisRes:
raise ValueError(
"The Pillow ImageFont instance has not getsize or getlength func."
)
+
+ @staticmethod
+ def concat_imgs(
+ imgs: List[np.ndarray], direction: str = "horizontal"
+ ) -> np.ndarray:
+ img_list = [Image.fromarray(img_path) for img_path in imgs]
+
+ img_sizes = np.array([v.size for v in img_list])
+
+ if direction == "horizontal":
+ width = np.sum(img_sizes[:, 0])
+ height = np.max(img_sizes[:, 1])
+ elif direction == "vertical":
+ width = np.max(img_sizes[:, 0])
+ height = np.sum(img_sizes[:, 1])
+ else:
+ raise ValueError(f"{direction} is not supported.")
+
+ new_image = Image.new("RGB", (width, height), color="white")
+
+ for i, img in enumerate(img_list):
+ if i == 0:
+ new_image.paste(img, (0, 0))
+ continue
+
+ if direction == "horizontal":
+ x = np.sum(img_sizes[:, 0][:i])
+ new_image.paste(img, (x, 0))
+ elif direction == "vertical":
+ y = np.sum(img_sizes[:, 1][:i])
+ new_image.paste(img, (0, y))
+
+ return np.array(new_image)
diff --git a/python/tests/test_files/korean.jpg b/python/tests/test_files/korean.jpg
new file mode 100644
index 0000000..08c4784
Binary files /dev/null and b/python/tests/test_files/korean.jpg differ
diff --git a/python/tests/test_main.py b/python/tests/test_main.py
index b338d4b..499ea95 100644
--- a/python/tests/test_main.py
+++ b/python/tests/test_main.py
@@ -38,6 +38,34 @@ def get_engine(params: Optional[Dict[str, Any]] = None):
return engine
+def test_vis_only_det(engine):
+ img_path = tests_dir / "ch_en_num.jpg"
+ result = engine(img_path, use_det=True, use_cls=False, use_rec=False)
+ vis_img = result.vis()
+ assert vis_img.shape == (430, 323, 3)
+
+
+def test_vis_only_rec(engine):
+ img_path = tests_dir / "text_rec.jpg"
+ result = engine(img_path, use_det=False, use_cls=False, use_rec=True)
+ vis_img = result.vis()
+ assert vis_img.shape == (78, 724, 3)
+
+
+def test_vis_only_cls(engine):
+ img_path = tests_dir / "text_rec.jpg"
+ result = engine(img_path, use_det=False, use_cls=True, use_rec=False)
+ vis_img = result.vis()
+ assert vis_img.shape == (78, 724, 3)
+
+
+def test_vis_det_cls_rec(engine):
+ img_path = tests_dir / "ch_en_num.jpg"
+ result = engine(img_path, use_det=False, use_cls=True, use_rec=False)
+ vis_img = result.vis()
+ assert vis_img.shape == (430, 323, 3)
+
+
def test_full_black_img(engine):
img_path = tests_dir / "empty_black.jpg"
result = engine(img_path)
@@ -92,7 +120,15 @@ def test_error_lang():
assert exc_info.type is ValueError
-def test_lang():
+def test_korean_lang():
+ engine = get_engine(params={"Global.lang_rec": "korean_mobile"})
+ img_path = tests_dir / "korean.jpg"
+ result = engine(img_path)
+ assert result.txts is not None
+ assert result.txts[0] == "베이징차오양"
+
+
+def test_en_lang():
engine = get_engine(params={"Global.lang_rec": "en_mobile"})
img_path = tests_dir / "en.jpg"
result = engine(img_path)