Seal Text Detection Module Tutorial¶

I. Overview¶

The seal text detection module typically outputs multi-point bounding boxes around text regions, which are then passed as inputs to the distortion correction and text recognition modules for subsequent processing to identify the textual content of the seal. Recognizing seal text is an integral part of document processing and finds applications in various scenarios such as contract comparison, inventory access auditing, and invoice reimbursement verification. The seal text detection module serves as a subtask within OCR (Optical Character Recognition), responsible for locating and marking the regions containing seal text within an image. The performance of this module directly impacts the accuracy and efficiency of the entire seal text OCR system.

II. Supported Model List¶

Model Name	Model Download Link	Hmean（%）	GPU Inference Time (ms) [Normal Mode / High-Performance Mode]	CPU Inference Time (ms) [Normal Mode / High-Performance Mode]	Model Size (M)	Description
PP-OCRv4_server_seal_det	Inference Model/Training Model	98.21	74.75 / 67.72	382.55 / 382.55	109 M	The server-side seal text detection model of PP-OCRv4 boasts higher accuracy and is suitable for deployment on better-equipped servers.
PP-OCRv4_mobile_seal_det	Inference Model/Training Model	96.47	7.82 / 3.09	48.28 / 23.97	4.6 M	The mobile-side seal text detection model of PP-OCRv4, on the other hand, offers greater efficiency and is suitable for deployment on end devices.

Test Environment Description:

Performance Test Environment
- Test Dataset： PaddleX Custom Dataset, Containing 500 Images of Circular Stamps.
- Hardware Configuration：
  - GPU: NVIDIA Tesla T4
  - CPU: Intel Xeon Gold 6271C @ 2.60GHz
  - Other Environments: Ubuntu 20.04 / cuDNN 8.6 / TensorRT 8.5.2.2
Inference Mode Description

Mode	GPU Configuration	CPU Configuration	Acceleration Technology Combination
Normal Mode	FP32 Precision / No TRT Acceleration	FP32 Precision / 8 Threads	PaddleInference
High-Performance Mode	Optimal combination of pre-selected precision types and acceleration strategies	FP32 Precision / 8 Threads	Pre-selected optimal backend (Paddle/OpenVINO/TRT, etc.)

III. Quick Integration ¶

❗ Before quick integration, please install the PaddleOCR wheel package. For detailed instructions, refer to PaddleOCR Local Installation Tutorial。

Quickly experience with just one command:

paddleocr seal_text_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png

You can also integrate the model inference from the layout area detection module into your project. Before running the following code, please download Example Image Go to the local area.

from paddleocr import SealTextDetection
model = SealTextDetection(model_name="PP-OCRv4_server_seal_det")
output = model.predict("seal_text_det.png", batch_size=1)
for res in output:
    res.print()
    res.save_to_img(save_path="./output/")
    res.save_to_json(save_path="./output/res.json")

After running, the result is:

{'res': {'input_path': 'seal_text_det.png', 'page_index': None, 'dt_polys': [array([[463, 477],
       ...,
       [428, 505]]), array([[297, 444],
       ...,
       [230, 443]]), array([[457, 346],
       ...,
       [267, 345]]), array([[325,  38],
       ...,
       [322,  37]])], 'dt_scores': [0.9912680344777314, 0.9906849624837963, 0.9847219455533163, 0.9914791724153904]}}

The meanings of the parameters are as follows: - input_path: represents the path of the input image to be predicted - dt_polys: represents the predicted text detection boxes, where each text detection box contains multiple vertices of a polygon. Each vertex is a list of two elements, representing the x and y coordinates of the vertex respectively - dt_scores: represents the confidence scores of the predicted text detection boxes

The visualization image is as follows:

Visualization Image

The explanations of related methods and parameters are as follows:

SealTextDetection instantiates a text detection model (here we take PP-OCRv4_server_seal_det as an example), and the specific explanations are as follows:

Parameter	Parameter Description	Parameter Type	Options	Default Value
`model_name`	Name of the model	`str`	All model names supported by PaddleX for seal text detection	None
`model_dir`	Path to store the model	`str`	None	None
`device`	The device used for model inference	`str`	It supports specifying specific GPU card numbers, such as "gpu:0", other hardware card numbers, such as "npu:0", or CPU, such as "cpu".	`gpu:0`
`limit_side_len`	Limit on the side length of the image for detection	`int/None`	int: Any integer greater than 0 None: If set to None, the default value from the official PaddleX model configuration will be used	None
`limit_type`	Type of side length limit for detection	`str/None`	str: Supports min and max. min ensures the shortest side of the image is not less than det_limit_side_len, max ensures the longest side is not greater than limit_side_len None: If set to None, the default value from the official PaddleX model configuration will be used	None
`thresh`	In the output probability map, pixels with scores greater than this threshold will be considered as text pixels	`float/None`	float: Any float greater than 0 None: If set to None, the default value from the official PaddleX model configuration will be used	None
`box_thresh`	If the average score of all pixels within a detection result box is greater than this threshold, the result will be considered as a text region	`float/None`	float: Any float greater than 0 None: If set to None, the default value from the official PaddleX model configuration will be used	None
`max_candidates`	Maximum number of text boxes to output	`int/None`	int: Any integer greater than 0 None: If set to None, the default value from the official PaddleX model configuration will be used	None
`unclip_ratio`	Expansion ratio for the Vatti clipping algorithm, used to expand the text region	`float/None`	float: Any float greater than 0 None: If set to None, the default value from the official PaddleX model configuration will be used	None
`use_dilation`	Whether to dilate the segmentation result	`bool/None`	True/False/None	None
`use_hpip`	Whether to enable the high-performance inference plugin	`bool`	None	`False`
`hpi_config`	High-performance inference configuration	`dict` \| `None`	None	`None`

The model_name must be specified. After specifying model_name, the built-in model parameters of PaddleX will be used by default. On this basis, if model_dir is specified, the user-defined model will be used.
The predict() method of the seal text detection model is called for inference prediction. The parameters of the predict() method include input, batch_size, limit_side_len, limit_type, thresh, box_thresh, max_candidates, unclip_ratio, and use_dilation. The specific descriptions are as follows:

Parameter	Parameter Description	Parameter Type	Options	Default Value
`input`	Data to be predicted, supporting multiple input types	`Python Var`/`str`/`dict`/`list`	Python Variable, such as image data represented by `numpy.ndarray` File Path, such as the local path of an image file: `/root/data/img.jpg` URL Link, such as the web URL of an image file: Example Local Directory, the directory should contain the data files to be predicted, such as the local path: `/root/data/` List, the elements of the list should be of the above-mentioned data types, such as `[numpy.ndarray, numpy.ndarray]`, `[\"/root/data/img1.jpg\", \"/root/data/img2.jpg\"]`, `[\"/root/data1\", \"/root/data2\"]`	None
`batch_size`	Batch size	`int`	Any integer greater than 0	1
`limit_side_len`	Side length limit for detection	`int/None`	int: Any integer greater than 0 None: If set to None, the parameter value initialized by the model will be used by default	None
`limit_type`	Type of side length limit for detection	`str/None`	str: Supports min and max. min indicates that the shortest side of the image is not less than det_limit_side_len, max indicates that the longest side of the image is not greater than limit_side_len None: If set to None, the parameter value initialized by the model will be used by default	None
`thresh`	In the output probability map, pixels with scores greater than this threshold will be considered as text pixels	`float/None`	float: Any float greater than 0 None: If set to None, the parameter value initialized by the model will be used by default	None
`box_thresh`	If the average score of all pixels within the detection result box is greater than this threshold, the result will be considered as a text area	`float/None`	float: Any float greater than 0 None: If set to None, the parameter value initialized by the model will be used by default	None
`max_candidates`	Maximum number of text boxes to be output	`int/None`	int: Any integer greater than 0 None: If set to None, the parameter value initialized by the model will be used by default	None
`unclip_ratio`	Expansion coefficient of the Vatti clipping algorithm, used to expand the text area	`float/None`	float: Any float greater than 0 None: If set to None, the parameter value initialized by the model will be used by default	None
`use_dilation`	Whether to dilate the segmentation result	`bool/None`	True/False/None	None

Process the prediction results. Each sample's prediction result is a corresponding Result object, and it supports operations such as printing, saving as an image, and saving as a json file:

Method	Method Description	Parameter	Parameter Type	Parameter Description	Default Value
`print()`	Print the result to the terminal	`format_json`	`bool`	Whether to format the output content using `JSON` indentation	`True`
		`indent`	`int`	Specify the indentation level to beautify the output `JSON` data, making it more readable. This is only effective when `format_json` is `True`	4
		`ensure_ascii`	`bool`	Control whether to escape non-`ASCII` characters to `Unicode`. When set to `True`, all non-`ASCII` characters will be escaped; `False` retains the original characters. This is only effective when `format_json` is `True`	`False`
`save_to_json()`	Save the result as a file in JSON format	`save_path`	`str`	The file path for saving. When it is a directory, the saved file name will be consistent with the input file name	None
		`indent`	`int`	Specify the indentation level to beautify the output `JSON` data, making it more readable. This is only effective when `format_json` is `True`	4
		`ensure_ascii`	`bool`	Control whether to escape non-`ASCII` characters to `Unicode`. When set to `True`, all non-`ASCII` characters will be escaped; `False` retains the original characters. This is only effective when `format_json` is `True`	`False`
`save_to_img()`	Save the result as a file in image format	`save_path`	`str`	The file path for saving. When it is a directory, the saved file name will be consistent with the input file name	None

In addition, it also supports obtaining visual images with results and prediction results through attributes, as follows:

Attribute	Attribute Description
`json`	Get the prediction result in `json` format
`img`	Get the visual image in `dict` format

IV. Custom Development¶

If the above model is still not performing well in your scenario, you can try the following steps for secondary development. Here, we'll use training PP-OCRv4_server_seal_det as an example; you can replace it with the corresponding configuration files for other models. First, you need to prepare a text detection dataset. You can refer to the format of the seal text detection demo data for preparation. Once prepared, you can follow the steps below for model training and export. After export, you can quickly integrate the model into the above API. This example uses a seal text detection demo dataset. Before training the model, please ensure that you have installed the dependencies required by PaddleOCR as per the installation documentation.

4.1 Dataset and Pre-trained Model Preparation¶

4.1.1 Preparing the Dataset¶

wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_curve_det_dataset_examples.tar -P ./dataset
tar -xf ./dataset/ocr_curve_det_dataset_examples.tar -C ./dataset/

4.1.1 Preparing the pre-trained model¶

wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams

4.2 Model Training¶

PaddleOCR has modularized the code, and when training the PP-OCRv4_server_seal_det model, you need to use the configuration file for PP-OCRv4_server_seal_det.

The training commands are as follows:

# Single GPU training (default training method)
python3 tools/train.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml \
   -o Global.pretrained_model=./PP-OCRv4_server_seal_det_pretrained.pdparams

# Multi-GPU training, specify GPU ids using the --gpus parameter
python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml \
        -o Global.pretrained_model=./PP-OCRv4_server_seal_det_pretrained.pdparams

4.3 Model Evaluation¶

You can evaluate the trained weights, such as output/xxx/xxx.pdparams, using the following command:

# Make sure to set the pretrained_model path to the local path. If using a model that was trained and saved by yourself, be sure to modify the path and filename to {path/to/weights}/{model_name}.
# Demo test set evaluation
python3 tools/eval.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml -o \
Global.pretrained_model=output/xxx/xxx.pdparams

4.4 Model Export¶

python3 tools/export_model.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml -o \
Global.pretrained_model=output/xxx/xxx.pdparams \
save_inference_dir="./PP-OCRv4_server_seal_det_infer/"

After exporting the model, the static graph model will be stored in the ./PP-OCRv4_server_seal_det_infer/ directory. In this directory, you will see the following files:

./PP-OCRv4_server_seal_det_infer/
├── inference.json
├── inference.pdiparams
├── inference.yml

With this, the secondary development is complete, and the static graph model can be directly integrated into PaddleOCR's API.