Skip to content

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

Update python openvino

SWHLf6bb96ca1d4e
diff --git a/python/README.md b/python/README.md
index 356e479..c95a823 100755
--- a/python/README.md
+++ b/python/README.md
@@ -4,8 +4,9 @@
### TODO
- [ ] 模型转INT8,正在尝试
-- [ ] 其他推理模型代码整理
-- [ ] 模型转换代码整理
+- [ ] OpenVINO推理方向分类模型有误,已经提了[issue](https://github.com/openvinotoolkit/openvino/issues/11501)
+- [x] 其他推理模型代码整理
+- [x] 模型转换代码整理
### 关于OpenVINO
- OpenVINO可以直接推理IR、ONNX和PaddlePaddle模型,具体如下(图来源:[link](https://docs.openvino.ai/latest/openvino_docs_OV_UG_OV_Runtime_User_Guide.html#doxid-openvino-docs-o-v-u-g-o-v-runtime-user-guide)):
diff --git a/python/onnxruntime_infer/ch_ppocr_mobile_v2_cls/text_cls.py b/python/onnxruntime_infer/ch_ppocr_mobile_v2_cls/text_cls.py
index e66b84f..6c706cd 100644
--- a/python/onnxruntime_infer/ch_ppocr_mobile_v2_cls/text_cls.py
+++ b/python/onnxruntime_infer/ch_ppocr_mobile_v2_cls/text_cls.py
@@ -106,9 +106,9 @@ class TextClassifier(object):
norm_img = norm_img[np.newaxis, :]
norm_img_batch.append(norm_img)
norm_img_batch = np.concatenate(norm_img_batch)
- onnx_inputs = {self.session.get_inputs()[0].name: norm_img_batch}
starttime = time.time()
+ onnx_inputs = {self.session.get_inputs()[0].name: norm_img_batch}
prob_out = self.session.run(None, onnx_inputs)[0]
cls_result = self.postprocess_op(prob_out)
diff --git a/python/onnxruntime_infer/requirements.txt b/python/onnxruntime_infer/requirements.txt
index 3bccfbb..fe8d176 100644
--- a/python/onnxruntime_infer/requirements.txt
+++ b/python/onnxruntime_infer/requirements.txt
@@ -1,8 +1,7 @@
pyclipper==1.2.0
Shapely==1.7.1
-onnxruntime==1.7.0
+onnxruntime>=1.7.0
opencv_python==4.5.1.48
six==1.15.0
-numpy==1.19.2
-Pillow==8.3.0
-openvino==2022.1.0
\ No newline at end of file
+numpy>=1.19.2
+Pillow==8.3.0
\ No newline at end of file
diff --git a/python/openvino_infer/README.md b/python/openvino_infer/README.md
index 7310bd2..cc8d650 100644
--- a/python/openvino_infer/README.md
+++ b/python/openvino_infer/README.md
@@ -7,16 +7,16 @@ pip install openvino==2022.1.0
pip install openvino-dev==2022.1.0
```
-#### 模型转换为IR
-```bash
-# 转动态模型 推理时占用内存平均为3.5G左右
-mo --input_model models/ch_PP-OCRv2_det_infer.onnx --output_dir models/IR/
+#### 模型问题
+- 因为OpenVINO可以直接推理ONNX模型,故这里暂时不作转换,直接推理之前ONNX模型即可
+- 这里仍然给出转换的代码,用作参考:
+ ```bash
+ mo --input_model models/ch_PP-OCRv2_det_infer.onnx --output_dir models/IR/
-# 转静态模型 推理时占用内存平均为1G左右
-mo --input_model models/ch_PP-OCRv2_det_infer.onnx \
- --output_dir models/IR/static \
- --input_shape "[1,3,12128,800]"
-```
+ mo --input_model models/ch_PP-OCRv2_det_infer.onnx \
+ --output_dir models/IR/static \
+ --input_shape "[1,3,12128,800]"
+ ```
-#### 测试图像下载
-- [long1.jpg](https://drive.google.com/file/d/1iJcGvOVIdUlyOS52bBdvO8uzx8QORo5M/view?usp=sharing)
+#### 使用步骤
+同[onnxruntime_infer](../onnxruntime_infer/README.md)
\ No newline at end of file
diff --git a/python/openvino_infer/ch_ppocr_mobile_v2_cls/text_cls.py b/python/openvino_infer/ch_ppocr_mobile_v2_cls/text_cls.py
index e882814..b5f02e0 100644
--- a/python/openvino_infer/ch_ppocr_mobile_v2_cls/text_cls.py
+++ b/python/openvino_infer/ch_ppocr_mobile_v2_cls/text_cls.py
@@ -18,6 +18,7 @@ import time
import cv2
import numpy as np
+import onnxruntime as ort
from openvino.runtime import Core
try:
@@ -33,10 +34,15 @@ class TextClassifier(object):
self.cls_thresh = 0.9
self.postprocess_op = ClsPostProcess(label_list=['0', '180'])
- ie = Core()
- model_onnx = ie.read_model(cls_model_path)
- compile_model = ie.compile_model(model=model_onnx, device_name='CPU')
- self.session = compile_model.create_infer_request()
+ # ie = Core()
+ # model_onnx = ie.read_model(cls_model_path)
+ # compile_model = ie.compile_model(model=model_onnx, device_name='CPU')
+ # self.session = compile_model.create_infer_request()
+
+ sess_opt = ort.SessionOptions()
+ sess_opt.log_severity_level = 4
+ sess_opt.enable_cpu_mem_arena = False
+ self.session = ort.InferenceSession(cls_model_path, sess_opt)
def resize_norm_img(self, img):
imgC, imgH, imgW = self.cls_image_shape
@@ -110,8 +116,11 @@ class TextClassifier(object):
norm_img_batch = np.concatenate(norm_img_batch)
starttime = time.time()
- self.session.infer(inputs=[norm_img_batch])
- prob_out = self.session.get_output_tensor().data
+ onnx_inputs = {self.session.get_inputs()[0].name: norm_img_batch}
+ prob_out = self.session.run(None, onnx_inputs)[0]
+
+ # self.session.infer(inputs=[norm_img_batch])
+ # prob_out = self.session.get_output_tensor().data
cls_result = self.postprocess_op(prob_out)
elapse += time.time() - starttime
diff --git a/python/openvino_infer/requirements.txt b/python/openvino_infer/requirements.txt
new file mode 100644
index 0000000..b65da24
--- /dev/null
+++ b/python/openvino_infer/requirements.txt
@@ -0,0 +1,7 @@
+pyclipper==1.2.0
+Shapely==1.7.1
+opencv_python==4.5.1.48
+six==1.15.0
+numpy>=1.19.2
+Pillow==8.3.0
+openvino==2022.1.0
\ No newline at end of file
diff --git a/python/openvino_infer/test_demo.py b/python/openvino_infer/test_demo.py
index 70cf959..2f33cee 100644
--- a/python/openvino_infer/test_demo.py
+++ b/python/openvino_infer/test_demo.py
@@ -16,7 +16,7 @@ text_sys = TextSystem(det_model_path,
cls_model_path=cls_model_path,
keys_path=keys_path)
-image_path = r'test_images/det_images/ch_en_num.jpg'
+image_path = r'test_images/ch_en_num.jpg'
dt_boxes, rec_res = text_sys(image_path)
visualize(image_path, dt_boxes, rec_res)
diff --git a/python/openvino_infer/test_demo.sh b/python/openvino_infer/test_demo.sh
new file mode 100644
index 0000000..e905249
--- /dev/null
+++ b/python/openvino_infer/test_demo.sh
@@ -0,0 +1,16 @@
+#! /bin/bash
+
+det_model_path="models/ch_PP-OCRv2_det_infer.onnx"
+cls_model_path="models/ch_ppocr_mobile_v2.0_cls_infer.onnx"
+
+# 这里可以按照自己需求,做相关更改
+rec_model_path="models/ch_ppocr_mobile_v2.0_rec_infer.onnx"
+keys_path="ch_ppocr_mobile_v2_rec/ppocr_keys_v1.txt"
+
+img_path="test_images/det_images/ch_en_num.jpg"
+
+python rapid_ocr_api.py --det_model_path ${det_model_path} \
+ --cls_model_path ${cls_model_path} \
+ --rec_model_path ${rec_model_path} \
+ --image_path ${img_path} \
+ --keys_path ${keys_path}
diff --git a/python/openvino_infer/test_openvino.py b/python/openvino_infer/test_openvino.py
index bcb6725..0b78a60 100644
--- a/python/openvino_infer/test_openvino.py
+++ b/python/openvino_infer/test_openvino.py
@@ -4,6 +4,13 @@
# @Author: SWHL
# @Contact: [email protected]
import numpy as np
+from openvino.runtime import Core
+
+cls_model_path = r'models/ch_ppocr_mobile_v2.0_cls_infer.onnx'
+ie = Core()
+model_onnx = ie.read_model(cls_model_path)
+compile_model = ie.compile_model(model_onnx, device_name='CPU')
+pass
# Infer ONNXRuntime
# import cv2
@@ -32,17 +39,17 @@ import numpy as np
# dt_boxes, elapse = text_detect_vino(img)
# print(f'openvino ONNX: {elapse}\tdt_boxes:{np.mean(dt_boxes)}')
-# # Infer IR
-import cv2
-from ch_ppocr_mobile_v2_det_openvino_ir import TextDetector as TextDetectorIR
+# # # Infer IR
+# import cv2
+# from ch_ppocr_mobile_v2_det_openvino_ir import TextDetector as TextDetectorIR
-det_model_path = 'models/IR/static/ch_PP-OCRv2_det_infer.xml'
-img_path = 'test_images/long1.jpg'
-img = cv2.imread(img_path)
+# det_model_path = 'models/IR/static/ch_PP-OCRv2_det_infer.xml'
+# img_path = 'test_images/long1.jpg'
+# img = cv2.imread(img_path)
-text_detect_ir = TextDetectorIR(det_model_path)
-dt_boxes, elapse = text_detect_ir(img)
-print(f'FP16 openvino IR: {elapse}\tdt_boxes:{np.mean(dt_boxes)}')
+# text_detect_ir = TextDetectorIR(det_model_path)
+# dt_boxes, elapse = text_detect_ir(img)
+# print(f'FP16 openvino IR: {elapse}\tdt_boxes:{np.mean(dt_boxes)}')
# Infer paddle 有误
# RuntimeError: Can't SetBlob with name: x, because model input (shape={?,3,960,960}) and blob (shape=(1.3.12128.800)) are incompatible
@@ -56,4 +63,3 @@ print(f'FP16 openvino IR: {elapse}\tdt_boxes:{np.mean(dt_boxes)}')
# text_detect_ir = TextDetectorPaddle(det_model_path)
# dt_boxes, elapse = text_detect_ir(img)
# print(f'openvino IR: {elapse}')
-