如何使用不同推理引擎？

引言¶

在rapidocr>=3.0.0版本之后，rapidocr可以单独为文本检测、文本行方向分类和文本识别单独指定不同的推理引擎。

例如：文本检测使用ONNXRuntime，文本识别使用PaddlePaddle（params={"Rec.engine_type": EngineType.PADDLE}）。同时，不同版本的OCR也可以通过Det.ocr_version灵活指定。

rapidocr支持4种推理引擎（ONNXRuntime / OpenVINO / PaddlePaddle / PyTorch），推荐首先使用 ONNXRuntime CPU 版。默认为ONNXRuntime。

rapidocr是通过指定不同参数来选择使用不同的推理引擎的。当然，使用不同推理引擎的前提是事先安装好对应的推理引擎库，并确保安装正确。

使用ONNXRuntime¶

安装ONNXRuntime。推荐用CPU版的ONNXRuntime，GPU版不推荐在rapidocr中使用，相关原因参见：ONNXRuntime GPU推理。
1
pip install onnxruntime

ONNXRuntime作为默认推理引擎，无需显式指定即可使用。

CPUNPUDirectML

from rapidocr import RapidOCR

engine = RapidOCR()

img_url = "https://github.com/RapidAI/RapidOCR/blob/main/python/tests/test_files/ch_en_num.jpg?raw=true"
result = engine(img_url)
print(result)

result.vis('vis_result.jpg')

Tip

仅在rapidocr>=3.1.0中支持。ONNXRuntime官方相关文档：link

安装
1
pip install rapidocr onnxruntime-cann

使用

from rapidocr import RapidOCR

engine = RapidOCR(params={"EngineConfig.onnxruntime.use_cann": True})

img_url = "<https://img1.baidu.com/it/u=3619974146,1266987475&fm=253&fmt=auto&app=138&f=JPEG?w=500&h=516>"
result = engine(img_url)
print(result)

result.vis("vis_result.jpg")

Tip

DirectML仅能Windows 10 Build 18362及以上使用。 ONNXRuntime官方相关文档：link

安装

pip install rapidocr onnxruntime-directml

使用

from rapidocr import RapidOCR

engine = RapidOCR(params={"EngineConfig.onnxruntime.use_dml": True})

img_url = "<https://img1.baidu.com/it/u=3619974146,1266987475&fm=253&fmt=auto&app=138&f=JPEG?w=500&h=516>"
result = engine(img_url)
print(result)

result.vis("vis_result.jpg")

查看输出日志。下面日志中打印出了 Using engine_name: onnxruntime，则证明使用的推理引擎是ONNXRuntime。

[INFO] 2025-03-21 09:28:03,457 base.py:30: Using engine_name: onnxruntime
[INFO] 2025-03-21 09:28:03,553 utils.py:35: File already exists in /Users/joshuawang/projects/_self/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_det_infer.onnx
[INFO] 2025-03-21 09:28:03,767 base.py:30: Using engine_name: onnxruntime
[INFO] 2025-03-21 09:28:03,768 utils.py:35: File already exists in /Users/joshuawang/projects/_self/RapidOCR/python/rapidocr/models/ch_ppocr_mobile_v2.0_cls_infer.onnx
[INFO] 2025-03-21 09:28:03,861 base.py:30: Using engine_name: onnxruntime
[INFO] 2025-03-21 09:28:03,862 utils.py:35: File already exists in /Users/joshuawang/projects/_self/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_rec_infer.onnx

使用OpenVINO¶

安装OpenVINO。
1
pip install openvino

指定OpenVINO作为推理引擎。

from rapidocr import RapidOCR, EngineType

engine = RapidOCR(
    params={
        "Det.engine_type": EngineType.OPENVINO,
        "Cls.engine_type": EngineType.OPENVINO,
        "Rec.engine_type": EngineType.OPENVINO,
    }
)

img_url = "https://github.com/RapidAI/RapidOCR/blob/main/python/tests/test_files/ch_en_num.jpg?raw=true"
result = engine(img_url)
print(result)

result.vis('vis_result.jpg')

查看输出日志。下面日志中打印出了 Using engine_name: openvino，则证明使用的推理引擎是OpenVINO。

[INFO] 2025-03-21 09:28:03,457 base.py:30: Using engine_name: openvino
[INFO] 2025-03-21 09:28:03,553 utils.py:35: File already exists in /Users/joshuawang/projects/_self/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_det_infer.onnx
[INFO] 2025-03-21 09:28:03,767 base.py:30: Using engine_name: openvino
[INFO] 2025-03-21 09:28:03,768 utils.py:35: File already exists in /Users/joshuawang/projects/_self/RapidOCR/python/rapidocr/models/ch_ppocr_mobile_v2.0_cls_infer.onnx
[INFO] 2025-03-21 09:28:03,861 base.py:30: Using engine_name: openvino
[INFO] 2025-03-21 09:28:03,862 utils.py:35: File already exists in /Users/joshuawang/projects/_self/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_rec_infer.onnx

使用PaddlePaddle¶

安装PaddlePaddle。

参见PaddlePaddle官方安装文档 → 快速安装

大家可以根据实际情况，选择安装需要的版本。

指定PaddlePaddle作为推理引擎。

CPUGPUNPU

from rapidocr import EngineType, RapidOCR

engine = RapidOCR(
    params={
        "Det.engine_type": EngineType.PADDLE,
        "Cls.engine_type": EngineType.PADDLE,
        "Rec.engine_type": EngineType.PADDLE,
    }
)

img_url = "https://github.com/RapidAI/RapidOCR/blob/main/python/tests/test_files/ch_en_num.jpg?raw=true"
result = engine(img_url)
print(result)

result.vis('vis_result.jpg')

from rapidocr import EngineType, RapidOCR

engine = RapidOCR(
    params={
    "Det.engine_type": EngineType.PADDLE,
    "EngineConfig.paddle.use_cuda": True,  # 使用PaddlePaddle GPU版推理
    "EngineConfig.paddle.gpu_id": 0,  # 指定GPU id
    }
)

img_url = "https://github.com/RapidAI/RapidOCR/blob/main/python/tests/test_files/ch_en_num.jpg?raw=true"
result = engine(img_url)
print(result)

result.vis('vis_result.jpg')

Tip

仅在rapidocr>=3.3.0中支持。对应版本的PaddlePaddle安装文档：link

from rapidocr import EngineType, RapidOCR

engine = RapidOCR(
    params={
    "Det.engine_type": EngineType.PADDLE,
    "EngineConfig.paddle.use_npu": True,  # 使用PaddlePaddle NPU版推理
    "EngineConfig.paddle.npu_id": 0,  # 指定NPU id
    }
)

img_url = "https://github.com/RapidAI/RapidOCR/blob/main/python/tests/test_files/ch_en_num.jpg?raw=true"
result = engine(img_url)
print(result)

result.vis('vis_result.jpg')

查看输出日志。下面日志中打印出了 Using engine_name: paddle，则证明使用的推理引擎是PaddlePaddle。

[INFO] 2025-03-22 15:20:45,528 utils.py:35: File already exists in /Users/jiahuawang/projects/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_det_infer/inference.pdmodel
[INFO] 2025-03-22 15:20:45,529 utils.py:35: File already exists in /Users/jiahuawang/projects/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_det_infer/inference.pdiparams
[INFO] 2025-03-22 15:20:45,746 base.py:30: Using engine_name: paddle
[INFO] 2025-03-22 15:20:45,746 utils.py:35: File already exists in /Users/jiahuawang/projects/RapidOCR/python/rapidocr/models/ch_ppocr_mobile_v2_cls_infer/inference.pdmodel
[INFO] 2025-03-22 15:20:45,746 utils.py:35: File already exists in /Users/jiahuawang/projects/RapidOCR/python/rapidocr/models/ch_ppocr_mobile_v2_cls_infer/inference.pdiparams
[INFO] 2025-03-22 15:20:45,903 base.py:30: Using engine_name: paddle
[INFO] 2025-03-22 15:20:45,904 utils.py:35: File already exists in /Users/jiahuawang/projects/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_rec_infer/inference.pdmodel
[INFO] 2025-03-22 15:20:45,904 utils.py:35: File already exists in /Users/jiahuawang/projects/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_rec_infer/inference.pdiparams

使用PyTorch¶

安装PyTorch。

参见PyTorch官方安装文档 → Install PyTorch。

大家可以根据实际情况，选择安装需要的版本。

指定PyTorch作为推理引擎。

CPUGPUNPU

from rapidocr import RapidOCR

engine = RapidOCR(
    params={
        "Det.engine_type": EngineType.TORCH,
        "Cls.engine_type": EngineType.TORCH,
        "Rec.engine_type": EngineType.TORCH,
    }
)

img_url = "https://github.com/RapidAI/RapidOCR/blob/main/python/tests/test_files/ch_en_num.jpg?raw=true"
result = engine(img_url)
print(result)

result.vis('vis_result.jpg')

from rapidocr import EngineType, RapidOCR

engine = RapidOCR(
    params={
        "Det.engine_type": EngineType.TORCH,
        "Cls.engine_type": EngineType.TORCH,
        "Rec.engine_type": EngineType.TORCH,
        "EngineConfig.torch.use_cuda": True,  # 使用torch GPU版推理
        "EngineConfig.torch.gpu_id": 0,  # 指定GPU id
    }
)

img_url = "https://github.com/RapidAI/RapidOCR/blob/main/python/tests/test_files/ch_en_num.jpg?raw=true"
result = engine(img_url)
print(result)

result.vis('vis_result.jpg')

Tip

仅在rapidocr>3.4.0中支持。torch_npu官方相关文档：link

安装torch_npu，参见：docs。

使用

from rapidocr import EngineType, RapidOCR

engine = RapidOCR(
    params={
        "Det.engine_type": EngineType.TORCH,
        "Cls.engine_type": EngineType.TORCH,
        "Rec.engine_type": EngineType.TORCH,
        "EngineConfig.torch.use_npu": True,  # 使用torch NPU版推理
        "EngineConfig.torch.npu_id": 0,  # 指定NPU id
    }
)

img_url = "https://github.com/RapidAI/RapidOCR/blob/main/python/tests/test_files/ch_en_num.jpg?raw=true"
result = engine(img_url)
print(result)

result.vis('vis_result.jpg')

查看输出日志。下面日志中打印出了 Using engine_name: torch，则证明使用的推理引擎是PyTorch。

[INFO] 2025-03-22 15:39:13,241 base.py:30: Using engine_name: torch
[INFO] 2025-03-22 15:39:13,956 utils.py:35: File already exists in /Users/jiahuawang/projects/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_det_infer.pth
[INFO] 2025-03-22 15:39:14,136 base.py:30: Using engine_name: torch
[INFO] 2025-03-22 15:39:14,136 utils.py:35: File already exists in /Users/jiahuawang/projects/RapidOCR/python/rapidocr/models/ch_ptocr_mobile_v2.0_cls_infer.pth
[INFO] 2025-03-22 15:39:14,168 base.py:30: Using engine_name: torch
[INFO] 2025-03-22 15:39:14,168 utils.py:35: File already exists in /Users/jiahuawang/projects/RapidOCR/python/rapidocr/models/ch_PP-OCRv4_rec_infer.pth

如何使用不同推理引擎？

引言¶

使用ONNXRuntime¶

使用OpenVINO¶

使用PaddlePaddle¶

使用PyTorch¶

评论