Skip to content

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

feat: add PP-OCRv5 cls module

SWHLb73ec51ecfdd
diff --git a/python/rapidocr/ch_ppocr_cls/main.py b/python/rapidocr/ch_ppocr_cls/main.py
index ae9a2b9..ec31bdd 100644
--- a/python/rapidocr/ch_ppocr_cls/main.py
+++ b/python/rapidocr/ch_ppocr_cls/main.py
@@ -20,13 +20,19 @@ import cv2
import numpy as np
from rapidocr.inference_engine.base import get_engine
+from rapidocr.utils.typings import OCRVersion
from .utils import ClsPostProcess, TextClsOutput
+CLS_SHAPE_BY_OCR_VERSION = {
+ OCRVersion.PPOCRV4: [3, 48, 192],
+ OCRVersion.PPOCRV5: [3, 80, 160],
+}
+
class TextClassifier:
def __init__(self, cfg: Dict[str, Any]):
- self.cls_image_shape = cfg["cls_image_shape"]
+ self.cls_image_shape = CLS_SHAPE_BY_OCR_VERSION[cfg["ocr_version"]]
self.cls_batch_num = cfg["cls_batch_num"]
self.cls_thresh = cfg["cls_thresh"]
self.postprocess_op = ClsPostProcess(cfg["label_list"])
@@ -83,7 +89,9 @@ class TextClassifier:
else:
resized_w = int(math.ceil(img_h * ratio))
- resized_image = cv2.resize(img, (resized_w, img_h))
+ resized_image = cv2.resize(
+ img, (resized_w, img_h), interpolation=cv2.INTER_LINEAR
+ )
resized_image = resized_image.astype("float32")
if img_c == 1:
resized_image = resized_image / 255
diff --git a/python/rapidocr/inference_engine/base.py b/python/rapidocr/inference_engine/base.py
index 128dfc1..75c49f7 100644
--- a/python/rapidocr/inference_engine/base.py
+++ b/python/rapidocr/inference_engine/base.py
@@ -136,7 +136,7 @@ class InferSession(abc.ABC):
return model_dict[k]
for k in model_dict:
- if k.startswith(lang_type):
+ if k.startswith(lang_type) and model_type in k:
return model_dict[k]
logger.error(