Skip to content

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

[python]Move the dict into the inside of onnx.

SWHL683743935ca8
diff --git a/python/README.md b/python/README.md
index 7fea7c1..9f8bf16 100755
--- a/python/README.md
+++ b/python/README.md
@@ -24,27 +24,20 @@
- 每个独立的模块下均有独立的`config.yaml`配置文件,可以单独使用
- `det`部分:
- `det`中`mobile`和`server`版,推理代码一致,直接更改配置文件中模型路径即可
- - `det`中`v2`和`v3`两个版本,推理代码一致,直接更改配置文件中模型路径即可
- ```yaml
- Det:
- module_name: ch_ppocr_v2_det
- class_name: TextDetector
- model_path: resources/models/ch_PP-OCRv3_det_infer.onnx
- ```
+ - `det`中`v2`和`v3`两个版本,推理代码一致。
- `rec`部分:
- `rec`中`mobile`和`server`版本,推理代码一致,直接更改配置文件中模型路径即可
- `rec`中`v2`和`v3`两个版本,共用同一个推理代码。
- 两版本差别仅在输入shape和模型。经过测试,采用`v3 rec模型`+`[3, 48, 320]`效果最好。
- - 目前配置文件`config.yaml`中(如下所示)已经改为该最优组合。
- ```yaml
- module_name: ch_ppocr_v2_rec
- class_name: TextRecognizer
- model_path: resources/models/ch_PP-OCRv3_rec_infer.onnx
-
- rec_img_shape: [3, 48, 320]
- rec_batch_num: 6
- keys_path: resources/rec_dict/ppocr_keys_v1.txt
- ```
+ - 目前配置文件`config.yaml`中(如下所示)已是最优组合。
+ ```yaml
+ module_name: ch_ppocr_v3_rec
+ class_name: TextRecognizer
+ model_path: resources/models/ch_PP-OCRv3_rec_infer.onnx
+
+ rec_img_shape: [3, 48, 320]
+ rec_batch_num: 6
+ ```
- onnxruntime和openvino调用方式如下:
```python
# 基于onnxruntime引擎推理
@@ -102,15 +95,15 @@ print(rec_res)
```
### 源码使用步骤
-1. 下载当前下的`rapidocr_onnxruntime`/`rapidocr_openvino`目录到本地
+1. 下载当前目录下的`rapidocr_onnxruntime`或者`rapidocr_openvino`目录到本地
2. 下载链接下的`resources`目录(包含模型和显示的字体文件)
- 下载链接:[百度网盘](https://pan.baidu.com/s/1PTcgXG2zEgQU6A_A3kGJ3Q?pwd=jhai) | [Google Drive](https://drive.google.com/drive/folders/1x_a9KpCo_1blxH1xFOfgKVkw1HYRVywY?usp=sharing)
- `resources/models`下模型搭配已经为最优组合(速度和精度平衡)
```text
- ch_PP-OCRv3_det + ch_ppocr_mobile_v2.0_cls + ch_ppocr_mobile_v2.0_rec
+ ch_PP-OCRv3_det + ch_ppocr_mobile_v2.0_cls + ch_PP-OCRv3_rec
```
- - 最终目录如下:
+ - 最终目录如下,自行比对:
```text
.
├── README.md
@@ -119,26 +112,24 @@ print(rec_res)
├── rapidocr_onnxruntime
│   ├── __init__.py
│   ├── ch_ppocr_v2_cls
- │   ├── ch_ppocr_v2_det
- │   ├── ch_ppocr_v2_rec
+ │   ├── ch_ppocr_v3_det
+ │   ├── ch_ppocr_v3_rec
│   └── rapid_ocr_api.py
├── rapidocr_openvino
│   ├── __init__.py
│ ├── README.md
│   ├── ch_ppocr_v2_cls
- │   ├── ch_ppocr_v2_det
- │   ├── ch_ppocr_v2_rec
+ │   ├── ch_ppocr_v3_det
+ │   ├── ch_ppocr_v3_rec
│   └── rapid_ocr_api.py
├── requirements.txt
├── resources
│ ├── fonts
- │ │   └── msyh.ttc
- │ ├── models
- │ │   ├── ch_PP-OCRv3_det_infer.onnx
- │ │   ├── ch_ppocr_mobile_v2.0_cls_infer.onnx
- │ │   └── ch_PP-OCRv3_rec_infer.onnx
- │ └── rec_dict
- │ └── ppocr_keys_v1.txt
+ │ │   └── FZYTK.TTF
+ │ └── models
+ │    ├── ch_PP-OCRv3_det_infer.onnx
+ │    ├── ch_ppocr_mobile_v2.0_cls_infer.onnx
+ │    └── ch_PP-OCRv3_rec_infer.onnx
└── test_images
├── ch_en_num.jpg
└── single_line_text.jpg
@@ -176,11 +167,11 @@ print(rec_res)
# 基于openvino引擎推理
# from rapidocr_openvino import TextSystem
- config_path = 'config.yaml'
- text_sys = TextSystem(config_path)
+ text_sys = TextSystem('config.yaml')
image_path = r'test_images/det_images/ch_en_num.jpg'
img = cv2.imread(image_path)
+
dt_boxes, rec_res = text_sys(img)
print(rec_res)
```
@@ -229,7 +220,6 @@ print(rec_res)
| ------------: | :----------: | :-----: | :----------------------------------------------|
|`rec_img_shape`| - |`[3, 48, 320]`| 输入文本识别模型的图像Shape(CHW) |
|`rec_batch_num`| - | 6 | 批次推理的batch大小,一般采用默认值即可,太大并没有明显提速,效果还可能会差 |
- |`keys_path`| - | - | 文本识别模型推理所使用字典文件,始识别哪种类型文本而定(中英、日文等) |
### onnxruntime-gpu版推理配置
diff --git a/python/config.yaml b/python/config.yaml
index a63f028..5b84273 100644
--- a/python/config.yaml
+++ b/python/config.yaml
@@ -70,4 +70,3 @@ Rec:
rec_img_shape: [3, 48, 320]
rec_batch_num: 6
- keys_path: resources/rec_dict/ppocr_keys_v1.txt
diff --git a/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/config.yaml b/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/config.yaml
index 9e149d6..e2dab8c 100644
--- a/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/config.yaml
+++ b/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/config.yaml
@@ -10,4 +10,3 @@ CUDAExecutionProvider:
rec_img_shape: [3, 48, 320]
rec_batch_num: 6
-keys_path: resources/rec_dict/ppocr_keys_v1.txt
diff --git a/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py b/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py
index 8b8677c..392e659 100644
--- a/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py
+++ b/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py
@@ -18,7 +18,7 @@ class OrtInferSession(object):
cuda_ep = 'CUDAExecutionProvider'
cpu_ep = 'CPUExecutionProvider'
cpu_provider_options = {
- "arena_extend_strategy": "kSameAsRequested",
+ "arena_extend_strategy": "kSameAsRequested",
}
EP_list = []
diff --git a/python/rapidocr_openvino/ch_ppocr_v3_rec/config.yaml b/python/rapidocr_openvino/ch_ppocr_v3_rec/config.yaml
index c66fe52..a9e4c13 100644
--- a/python/rapidocr_openvino/ch_ppocr_v3_rec/config.yaml
+++ b/python/rapidocr_openvino/ch_ppocr_v3_rec/config.yaml
@@ -2,4 +2,3 @@ model_path: resources/models/ch_PP-OCRv3_rec_infer.onnx
rec_img_shape: [3, 32, 320]
rec_batch_num: 6
-keys_path: resources/rec_dict/ppocr_keys_v1.txt
diff --git a/python/resources/rec_dict/ppocr_keys_v1.txt b/python/rapidocr_openvino/ch_ppocr_v3_rec/ppocr_keys_v1.txt
similarity index 100%
rename from python/resources/rec_dict/ppocr_keys_v1.txt
rename to python/rapidocr_openvino/ch_ppocr_v3_rec/ppocr_keys_v1.txt
diff --git a/python/rapidocr_openvino/ch_ppocr_v3_rec/text_recognize.py b/python/rapidocr_openvino/ch_ppocr_v3_rec/text_recognize.py
index f96b761..c3b7e96 100644
--- a/python/rapidocr_openvino/ch_ppocr_v3_rec/text_recognize.py
+++ b/python/rapidocr_openvino/ch_ppocr_v3_rec/text_recognize.py
@@ -18,6 +18,7 @@ import argparse
import math
import time
from typing import List
+from pathlib import Path
import cv2
import numpy as np
@@ -32,7 +33,9 @@ class TextRecognizer(object):
def __init__(self, config):
self.rec_image_shape = config['rec_img_shape']
self.rec_batch_num = config['rec_batch_num']
- self.character_dict_path = config['keys_path']
+
+ dict_path = str(Path(__file__).parent / './ppocr_keys_v1.txt')
+ self.character_dict_path = config.get( 'keys_path', dict_path)
self.postprocess_op = CTCLabelDecode(self.character_dict_path)
openvino_instance = OpenVINOInferSession(config)
diff --git a/python/test_demo.py b/python/test_demo.py
index c6c8d7d..5878b2f 100644
--- a/python/test_demo.py
+++ b/python/test_demo.py
@@ -91,10 +91,10 @@ def visualize(image_path, boxes, rec_res, font_path="resources/fonts/FZYTK.TTF")
if __name__ == '__main__':
text_sys = TextSystem('config.yaml')
- image_path = r'test_images/ch_en_num.jpg'
+ image_path = 'test_images/ch_en_num.jpg'
img = cv2.imread(image_path)
dt_boxes, rec_res = text_sys(img)
print(rec_res)
visualize(image_path, dt_boxes, rec_res,
- font_path='resources/fonts/msyh.ttc')
+ font_path='resources/fonts/FZYTK.TTF')
diff --git a/python/tests/benchmark/config.yaml b/python/tests/benchmark/config.yaml
index 4b5b414..5b84273 100644
--- a/python/tests/benchmark/config.yaml
+++ b/python/tests/benchmark/config.yaml
@@ -1,11 +1,12 @@
Global:
text_score: 0.5
use_angle_cls: true
- print_verbose: false
+ print_verbose: true
min_height: 30
+ width_height_ratio: 8
Det:
- module_name: ch_ppocr_v2_det
+ module_name: ch_ppocr_v3_det
class_name: TextDetector
model_path: resources/models/ch_PP-OCRv3_det_infer.onnx
@@ -36,6 +37,7 @@ Det:
max_candidates: 1000
unclip_ratio: 1.6
use_dilation: true
+ score_mode: fast
Cls:
module_name: ch_ppocr_v2_cls
@@ -55,7 +57,7 @@ Cls:
label_list: ['0', '180']
Rec:
- module_name: ch_ppocr_v2_rec
+ module_name: ch_ppocr_v3_rec
class_name: TextRecognizer
model_path: resources/models/ch_PP-OCRv3_rec_infer.onnx
@@ -68,4 +70,3 @@ Rec:
rec_img_shape: [3, 48, 320]
rec_batch_num: 6
- keys_path: resources/rec_dict/ppocr_keys_v1.txt
diff --git a/python/tests/benchmark/config_gpu.yaml b/python/tests/benchmark/config_gpu.yaml
index c8d3156..eaca238 100644
--- a/python/tests/benchmark/config_gpu.yaml
+++ b/python/tests/benchmark/config_gpu.yaml
@@ -3,9 +3,10 @@ Global:
use_angle_cls: true
print_verbose: false
min_height: 30
+ width_height_ratio: 8
Det:
- module_name: ch_ppocr_v2_det
+ module_name: ch_ppocr_v3_det
class_name: TextDetector
model_path: resources/models/ch_PP-OCRv3_det_infer.onnx
@@ -55,7 +56,7 @@ Cls:
label_list: ['0', '180']
Rec:
- module_name: ch_ppocr_v2_rec
+ module_name: ch_ppocr_v3_rec
class_name: TextRecognizer
model_path: resources/models/ch_PP-OCRv3_rec_infer.onnx
@@ -68,4 +69,3 @@ Rec:
rec_img_shape: [3, 48, 320]
rec_batch_num: 6
- keys_path: resources/rec_dict/ppocr_keys_v1.txt
diff --git a/python/tests/test_rec.py b/python/tests/test_rec.py
index 69fef63..790162d 100644
--- a/python/tests/test_rec.py
+++ b/python/tests/test_rec.py
@@ -12,7 +12,6 @@ TextRecognizer = base.init_module(module_name, class_name)
yaml_path = base.root_dir / base.package_name / module_name / 'config.yaml'
config = base.read_yaml(str(yaml_path))
config['model_path'] = str(base.root_dir / config['model_path'])
-config['keys_path'] = str(base.root_dir / config['keys_path'])
text_rec = TextRecognizer(config)