X

Model Details

Model Description

Paddle Text detection is an efficient and flexible text detection system designed for real-world applications. It supports over 80 languages and is optimized for deployment in resource-constrained environments. It achieves high recognition performance on multi-language text, including complex scripts, while remaining efficient for real-time use. It also offers robust capabilities for detecting and recognizing text in challenging conditions such as low-quality images or varying text orientations.

Developed by: PaddlePaddle
Shared by:
Model type: Computer vision
License:
Resources for more information:

Training Details

Training Data

The model was pretrained on dataset then finetuned on different real-world datasets.

Testing Details

Metrics

Metrics are taken from the original . Tests were performed on the and datasets.

Dataset	Precision	Recall	F-score
MSRA-TD500	90.4	76.3	82.8
CTW1500	84.8	77.5	81.0

Technical Specifications

Input/Output Details

Input:
- Name: x
  - Info: 0-255 BGR un-normalized image.
Output:
- Name: output
  - Info: A segmentation mask over the entire image, each output value corresponds to the probability of the input pixel being part of text.

Model Architecture

Backbone: ResNet18, the backbone consists of 18 layers with residual connections.
Segmentation Head: , allowing the network to dynamically learn and optimize the binarization process during training. This head enhances the model’s ability to distinguish between text and background.

Throughput

Model variant: paddle-text-detection:544x960

• Input shape: [1, 3, 544, 960] • Output shape: [1, 1, 544, 960]

• GFLOPs: 3.569

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC2	FP16	4.85	N/A
RVC4	INT8	299.13	3.67

Model variant: paddle-text-detection:320x576

• Input shape: [1, 3, 320, 576] • Output shape: [1, 1, 320, 576]

• GFLOPs: 1.260

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC2	FP16	13.90	N/A
RVC4	INT8	545.08	3.17

Model variant: paddle-text-detection:256x256

• Input shape: [1, 3, 256, 256] • Output shape: [1, 1, 256, 256]

• GFLOPs: 0.448

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC2	FP16	37.41	N/A
RVC4	FP16	508.11	3.21

* Benchmarked with , using 2 threads (and the DSP runtime in balanced mode for RVC4).

* Parameters and FLOPs are obtained from the package.

Utilization

Models converted for RVC Platforms can be used for inference on OAK devices. DepthAI pipelines are used to define the information flow linking the device, inference model, and the output parser (as defined in model head(s)). Below, we present the most crucial utilization steps for the particular model. Please consult the docs for more information.

Install DAIv3 and depthai-nodes libraries:

pip install depthai
pip install depthai-nodes

Define model:

model_description = dai.NNModelDescription(
    "luxonis/paddle-text-detection:256x256"
)

nn = pipeline.create(ParsingNeuralNetwork).build(
    <CameraNode>, model_description
)

Inspect model head(s):

PPTextDetectionParser that outputs message.

Get parsed output(s):

while pipeline.isRuning():
    parser_output: ImgDetectionsExtended = parser_output_queue.get()

Example

You can quickly run the model using our script. It automatically downloads the model, creates a DepthAI pipeline, runs the inference, and displays the results using our DepthAI visualizer tool. To try it out, run:

python3 main.py \
    --model luxonis/paddle-text-detection:256x256

You can also try running the example.

PaddlePaddle implementation of DB Text detection model.
License	Apache 2.0 Commercial use
Downloads	2961
Tasks	Object Detection
Model Types	ONNX

Name	Version	Available For	Created At	Deploy
		RVC2, RVC4	About 1 year ago
		RVC2, RVC4	About 1 year ago
		RVC2, RVC4	About 1 year ago