Rapid Layout

https://rapidai.github.io/RapidLayout/v1.2.1/
Admin
Rapid Layout is a document layout analysis tool that detects and locates elements like titles,...
Tokens:6,270
Snippets:67
Trust Score:9.7
Update:5 months ago
Show doc for...
Tokens:
Raw
### Install RapidLayout and onnxruntime

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/installation?q=

Install the RapidLayout library and the onnxruntime package using pip. This is the primary installation command for users who want to perform Chinese layout analysis directly.

```bash
pip install rapid-layout onnxruntime

```

--------------------------------

### Install and Use OpenVINO

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/how_to_use_other_engine

Install RapidLayout and OpenVINO, then configure RapidLayout to use the OpenVINO engine. Device and thread configurations for OpenVINO can be found in the `engine_cfg.yaml` file.

```bash
pip install rapid-layout onnxruntime openvino
```

```python
from rapid_layout import EngineType, ModelType, RapidLayout

layout_engine = RapidLayout(
    model_type=ModelType.PP_LAYOUT_CDLA,
    engine_type=EngineType.OPENVINO,
)
img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")
```

--------------------------------

### Terminal Execution

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/usage?q=

Provides examples of how to run RapidLayout directly from the terminal.

```APIDOC
## Terminal Execution

RapidLayout can be executed directly from the command line for quick processing of images.

### Usage Examples

**Basic usage:**
```bash
rapid_layout test_images/layout.png
```

**With custom parameters:**
```bash
rapid_layout test_images/layout.png -m pp_layout_cdla --conf_thresh 0.5 --iou_thresh 0.5
```
```

--------------------------------

### Python Usage with Configuration Object

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/usage?q=

Initialize the RapidLayout engine using a configuration object (RapidLayoutInput) for a more structured setup, equivalent to specifying keyword arguments.

```python
from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput

cfg = RapidLayoutInput(
    model_type=ModelType.PP_LAYOUT_CDLA,
    engine_type=EngineType.ONNXRUNTIME,
    conf_thresh=0.5,
    iou_thresh=0.5,
)
layout_engine = RapidLayout(cfg=cfg)
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")

```

--------------------------------

### Install ONNX Runtime GPU

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/how_to_use_other_engine

Install the `onnxruntime-gpu` package to enable GPU acceleration with ONNX Runtime. Ensure compatibility with your CUDA version.

```bash
pip install rapid_layout
# 请确保 onnxruntime-gpu 与当前 GPU/CUDA 版本对应
# 参见 https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements
pip install onnxruntime-gpu
```

--------------------------------

### Use ONNX Runtime with CUDA

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/how_to_use_other_engine

Configure RapidLayout to use ONNX Runtime with CUDA acceleration. Specify the device ID for multi-GPU setups.

```python
from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput

cfg = RapidLayoutInput(
    model_type=ModelType.PP_LAYOUT_CDLA,
    engine_type=EngineType.ONNXRUNTIME,
    engine_cfg={"use_cuda": True, "cuda_ep_cfg": {"device_id": 0}},
)
layout_engine = RapidLayout(cfg=cfg)

img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")
```

--------------------------------

### Run RapidLayout with Configuration Object

Source: https://rapidai.github.io/RapidLayout/v1.2.1/quickstart?q=

Initialize RapidLayout using a configuration object (`RapidLayoutInput`) to specify model and engine settings.

```python
from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput

cfg = RapidLayoutInput(
    model_type=ModelType.PP_LAYOUT_CDLA,
    engine_type=EngineType.ONNXRUNTIME,
)
layout_engine = RapidLayout(cfg=cfg)
img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")

```

--------------------------------

### Python Script - Using Configuration Object

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/usage?q=

Illustrates using a configuration object (`RapidLayoutInput`) to initialize RapidLayout, providing an alternative way to set parameters.

```APIDOC
## Python Script - Using Configuration Object

This method uses a `RapidLayoutInput` object to encapsulate all configuration parameters, offering an equivalent way to initialize `RapidLayout` as specifying keyword arguments directly.

### Code Example
```python
from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput

cfg = RapidLayoutInput(
    model_type=ModelType.PP_LAYOUT_CDLA,
    engine_type=EngineType.ONNXRUNTIME,
    conf_thresh=0.5,
    iou_thresh=0.5,
)
layout_engine = RapidLayout(cfg=cfg)
img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")
```
```

--------------------------------

### Run RapidLayout with Default Settings

Source: https://rapidai.github.io/RapidLayout/v1.2.1/quickstart?q=

Execute RapidLayout using the default model (`pp_layout_cdla`) and engine (`onnxruntime`) by providing an image path.

```python
from rapid_layout import RapidLayout

layout_engine = RapidLayout()
img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")

```

--------------------------------

### Constructor Parameters (RapidLayout / RapidLayoutInput)

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/usage?q=

Details the parameters available for the `RapidLayout` and `RapidLayoutInput` constructors.

```APIDOC
## Constructor Parameters (RapidLayout / RapidLayoutInput)

The following parameters can be used when initializing `RapidLayout` or `RapidLayoutInput`:

| Parameter        | Type                     | Default Value   | Description                     |
|------------------|--------------------------|-----------------|---------------------------------|
| `model_type`     | ModelType / str          | `pp_layout_cdla`| Model type                      |
| `model_dir_or_path`| str / Path / None        | None            | Model path, resolves by `model_type` if not provided |
| `engine_type`    | EngineType / str         | `onnxruntime`   | Inference engine: `onnxruntime`, `openvino` |
| `engine_cfg`     | dict                     | `{}`            | Additional engine configuration |
| `conf_thresh`    | float                    | 0.5             | Box confidence threshold [0, 1]   |
| `iou_thresh`     | float                    | 0.5             | IoU threshold [0, 1]            |
```

--------------------------------

### Python Script - Specifying Model and Engine

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/usage?q=

Shows how to specify the model type and engine type using keyword arguments in the RapidLayout constructor.

```APIDOC
## Python Script - Specifying Model and Engine

This example demonstrates how to customize the `RapidLayout` by explicitly setting the `model_type`, `engine_type`, `conf_thresh`, and `iou_thresh` during initialization.

### Code Example
```python
from rapid_layout import EngineType, ModelType, RapidLayout

layout_engine = RapidLayout(
    model_type=ModelType.PP_LAYOUT_CDLA,
    engine_type=EngineType.ONNXRUNTIME,
    conf_thresh=0.5,
    iou_thresh=0.5,
)
img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")
```
```

--------------------------------

### Python Script - Default Usage

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/usage?q=

Demonstrates the default usage of RapidLayout with the pp_layout_cdla model and onnxruntime engine.

```APIDOC
## Python Script - Default Usage

This snippet shows the basic initialization and usage of the `RapidLayout` class without specifying any custom parameters. It uses the default model (`pp_layout_cdla`) and engine (`onnxruntime`).

### Code Example
```python
from rapid_layout import RapidLayout

layout_engine = RapidLayout()
img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")
```
```

--------------------------------

### Run RapidLayout with Custom Model and Engine

Source: https://rapidai.github.io/RapidLayout/v1.2.1/quickstart?q=

Specify the model type, engine type, and confidence threshold when initializing RapidLayout for custom inference configurations.

```python
from rapid_layout import EngineType, ModelType, RapidLayout

layout_engine = RapidLayout(
    model_type=ModelType.PP_LAYOUT_CDLA,
    engine_type=EngineType.ONNXRUNTIME,
    conf_thresh=0.5,
)
img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")

```

--------------------------------

### Run RapidLayout from Terminal

Source: https://rapidai.github.io/RapidLayout/v1.2.1/quickstart?q=

Execute RapidLayout directly from the terminal, optionally specifying the model and confidence threshold.

```bash
rapid_layout test_images/layout.png
rapid_layout test_images/layout.png -m pp_layout_cdla --conf_thresh 0.5

```

--------------------------------

### Terminal Command for Layout Analysis

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/usage?q=

Run RapidLayout directly from the terminal to analyze images. Basic usage requires the image path, while advanced usage allows specifying the model and thresholds.

```bash
rapid_layout test_images/layout.png

```

```bash
rapid_layout test_images/layout.png -m pp_layout_cdla --conf_thresh 0.5 --iou_thresh 0.5

```

--------------------------------

### Use ONNX Runtime with NPU (CANN)

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/how_to_use_other_engine

Configure RapidLayout to use ONNX Runtime with CANN (NPU) acceleration. Refer to `engine_cfg.yaml` for detailed configuration parameters.

```python
from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput

cfg = RapidLayoutInput(
    model_type=ModelType.PP_LAYOUT_CDLA,
    engine_type=EngineType.ONNXRUNTIME,
    engine_cfg={"use_cann": True, "cann_ep_cfg": {"device_id": 0}},
)
layout_engine = RapidLayout(cfg=cfg)

img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")
```

--------------------------------

### PP-DocLayoutV3 ONNX Conversion Error

Source: https://rapidai.github.io/RapidLayout/v1.2.1/blog/2026/02/10/support-PP-DocLayoutv2-v3

This traceback shows a much smaller discrepancy (1.57%) for the PP-DocLayoutV3 ONNX conversion compared to PP-DocLayoutV2, indicating better compatibility with the specified tolerance.

```text
AssertionError:
Not equal to tolerance rtol=0, atol=0.001

Mismatched elements: 33 / 2100 (1.57%)
Max absolute difference among violations: 1.
Max relative difference among violations: 0.01754386
ACTUAL: array([[2.200000e+01, 9.658169e-01, 3.387792e+01, ..., 3.626684e+02,
        8.528884e+02, 1.540000e+02],
       [2.200000e+01, 9.657925e-01, 3.363610e+01, ..., 3.633332e+02,...
 DESIRED: array([[2.200000e+01, 9.658167e-01, 3.387791e+01, ..., 3.626685e+02,
        8.528885e+02, 1.530000e+02],
       [2.200000e+01, 9.657924e-01, 3.363615e+01, ..., 3.633333e+02,... 
```

--------------------------------

### PP-DocLayoutV2 ONNX Conversion Error

Source: https://rapidai.github.io/RapidLayout/v1.2.1/blog/2026/02/10/support-PP-DocLayoutv2-v3

This traceback indicates a significant difference (14.8%) between the Paddle model and its ONNX conversion for PP-DocLayoutV2, exceeding the allowed tolerance of 0.001. This issue has been reported.

```text
AssertionError:
Not equal to tolerance rtol=0, atol=0.001

Mismatched elements: 354 / 2400 (14.8%)
Max absolute difference: 196.
Max relative difference: 194.
 x: array([[2.200000e+01, 9.889924e-01, 3.354079e+01, ..., 6.150450e+02,
        2.900000e+02, 2.900000e+02],
       [2.200000e+01, 9.888635e-01, 3.372379e+01, ..., 8.526023e+02,...
 y: array([[2.200000e+01, 9.889925e-01, 3.354081e+01, ..., 6.150450e+02,
        2.900000e+02, 2.900000e+02],
       [2.200000e+01, 9.888635e-01, 3.372382e+01, ..., 8.526024e+02,... 
```

--------------------------------

### Convert PP-DocLayoutV2 Model to ONNX

Source: https://rapidai.github.io/RapidLayout/v1.2.1/blog/2026/02/10/support-PP-DocLayoutv2-v3?q=

Use this command to convert the PP-DocLayoutV2 model from PaddlePaddle format to ONNX format. Ensure the model directory and filenames are correct.

```bash
paddle2onnx  --model_dir=models/PP-DocLayoutV2  --model_filename inference.json --params_filename inference.pdiparams  --save_file=./models/PP-DocLayoutV2/inference.onnx  --enable_onnx_checker=True
```

--------------------------------

### Python Usage with Specified Model and Engine

Source: https://rapidai.github.io/RapidLayout/v1.2.1/install_usage/usage?q=

Configure the layout engine by specifying the model type, inference engine, confidence threshold, and IoU threshold using keyword arguments.

```python
from rapid_layout import EngineType, ModelType, RapidLayout

layout_engine = RapidLayout(
    model_type=ModelType.PP_LAYOUT_CDLA,
    engine_type=EngineType.ONNXRUNTIME,
    conf_thresh=0.5,
    iou_thresh=0.5,
)
results = layout_engine(img_path)
print(results)
results.vis("layout_res.png")

```

--------------------------------

### PP-DocLayoutV2 ONNX vs Paddle Comparison Error

Source: https://rapidai.github.io/RapidLayout/v1.2.1/blog/2026/02/10/support-PP-DocLayoutv2-v3?q=

This is a sample traceback showing the AssertionError when comparing ONNX and Paddle model outputs for PP-DocLayoutV2. It indicates a significant difference (14.8%) in the results, exceeding the specified tolerance.

```text
AssertionError:
Not equal to tolerance rtol=0, atol=0.001

Mismatched elements: 354 / 2400 (14.8%)
Max absolute difference: 196.
Max relative difference: 194.
 x: array([[2.200000e+01, 9.889924e-01, 3.354079e+01, ..., 6.150450e+02,
        2.900000e+02, 2.900000e+02],
       [2.200000e+01, 9.888635e-01, 3.372379e+01, ..., 8.526023e+02,... 
y: array([[2.200000e+01, 9.889925e-01, 3.354081e+01, ..., 6.150450e+02,
        2.900000e+02, 2.900000e+02],
       [2.200000e+01, 9.888635e-01, 3.372382e+01, ..., 8.526024e+02,... 
```

--------------------------------

### Compare ONNX and Paddle Inference for PP-DocLayoutV2

Source: https://rapidai.github.io/RapidLayout/v1.2.1/blog/2026/02/10/support-PP-DocLayoutv2-v3?q=

This Python snippet compares the inference results of an ONNX model against the original Paddle model for PP-DocLayoutV2. It uses numpy.testing.assert_allclose to check for discrepancies within a specified tolerance.

```python
# 省略前面代码... ...

import onnxruntime
import numpy as np

model_path = "models/PP-DocLayoutV2/inference.onnx"
ort_session = onnxruntime.InferenceSession(model_path)
ort_inputs = {
    "im_shape": batch_inputs[0],
    "image": batch_inputs[1],
    "scale_factor": batch_inputs[2],
}
ort_outputs = ort_session.run(None, ort_inputs)

# do infer
batch_preds = self.infer(batch_inputs)

# 千分位是否相同
np.testing.assert_allclose(batch_preds[0], ort_outputs[0], atol=1e-3, rtol=0)
```

--------------------------------

### PP-DocLayoutV3 ONNX vs Paddle Comparison Error

Source: https://rapidai.github.io/RapidLayout/v1.2.1/blog/2026/02/10/support-PP-DocLayoutv2-v3?q=

This traceback shows the AssertionError for PP-DocLayoutV3, indicating a much smaller difference (1.57%) between ONNX and Paddle model inference results compared to PP-DocLayoutV2.

```text
AssertionError:
Not equal to tolerance rtol=0, atol=0.001

Mismatched elements: 33 / 2100 (1.57%)
Max absolute difference among violations: 1.
Max relative difference among violations: 0.01754386
ACTUAL: array([[2.200000e+01, 9.658169e-01, 3.387792e+01, ..., 3.626684e+02,
        8.528884e+02, 1.540000e+02],
       [2.200000e+01, 9.657925e-01, 3.363610e+01, ..., 3.633332e+02,... 
 DESIRED: array([[2.200000e+01, 9.658167e-01, 3.387791e+01, ..., 3.626685e+02,
        8.528885e+02, 1.530000e+02],
       [2.200000e+01, 9.657924e-01, 3.363615e+01, ..., 3.633333e+02,... 
```
Tokens:
Raw