Skip to content

Model List

Introduction

For the commonly used models already released by PaddleOCR, we have carried out a unified conversion and aggregation here, including the PP-OCRv4, PP-OCRv5 and PP-OCRv6 series in PaddlePaddle, ONNX, MNN and PyTorch formats. TensorRT dynamically builds an Engine file matching the current hardware from the ONNX model on first run.

All models are currently hosted on ModelScope.

rapidocr v3 has integrated all the hosted models, which can be downloaded automatically by specifying the parameters below. Corresponding configuration file: default_models.yaml. Of course, you can also download them yourself from the link above.

Starting from rapidocr>=3.10.0, model selection is managed uniformly by the model routes in default_models.yaml. After the user passes in the language, model type and OCR version, RapidOCR automatically resolves the actual model file.

Starting from rapidocr>=3.10.0, lang_type is no longer limited to the LangDet, LangCls or LangRec enums; it can also be passed directly as a language code string, such as "japan", "korean" or "arabic". The string code must match a route for the current task, OCR version and model type; the supported codes and aliases are governed by the model route details in this document and default_models.yaml. The original enum syntax remains compatible.

For example, the following recognition configuration uses a language code string directly, without first converting it to the LangRec enum:

from rapidocr import EngineType, ModelType, OCRVersion, RapidOCR

engine = RapidOCR(
    params={
        "Rec.engine_type": EngineType.ONNXRUNTIME,
        "Rec.lang_type": "japan",
        "Rec.model_type": ModelType.MOBILE,
        "Rec.ocr_version": OCRVersion.PPOCRV4,
    }
)

When using RTL languages such as Arabic, please first install the RTL extra dependencies.

Default Configuration

After installing rapidocr via pip, you can use it directly without specifying any parameters, because the whl package pre-packages the default models and provides the default configuration at the same time.

The default configuration differs between versions of rapidocr. The default configurations corresponding to the different versions are given in detail below.

1
2
3
4
5
6
7
8
9
from rapidocr import RapidOCR

engine = RapidOCR()

img_url = "https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/test_files/ch_en_num.jpg"
result = engine(img_url)
print(result)

result.vis("vis_result.jpg")

Equivalent to the following:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
from rapidocr import EngineType, LangDet, LangRec, ModelType, OCRVersion, RapidOCR

engine = RapidOCR(
    params={
        "Det.engine_type": EngineType.ONNXRUNTIME,
        "Det.lang_type": LangDet.CH,
        "Det.model_type": ModelType.SMALL,
        "Det.ocr_version": OCRVersion.PPOCRV6,
        "Rec.engine_type": EngineType.ONNXRUNTIME,
        "Rec.lang_type": LangRec.CH,
        "Rec.model_type": ModelType.SMALL,
        "Rec.ocr_version": OCRVersion.PPOCRV6,
        "Cls.engine_type": EngineType.ONNXRUNTIME,
        "Cls.lang_type": LangDet.CH,
        "Cls.model_type": ModelType.MOBILE,
        "Cls.ocr_version": OCRVersion.PPOCRV4,
    }
)

img_url = "https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/test_files/ch_en_num.jpg"
result = engine(img_url)
print(result)

result.vis("vis_result.jpg")

Corresponding configuration file: config.yaml

1
2
3
4
5
6
7
8
9
from rapidocr import RapidOCR

engine = RapidOCR()

img_url = "https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/test_files/ch_en_num.jpg"
result = engine(img_url)
print(result)

result.vis("vis_result.jpg")

Equivalent to the following:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
from rapidocr import EngineType, LangDet, LangRec, ModelType, OCRVersion, RapidOCR

engine = RapidOCR(
    params={
        "Det.engine_type": EngineType.ONNXRUNTIME,
        "Det.lang_type": LangDet.CH,
        "Det.model_type": ModelType.MOBILE,
        "Det.ocr_version": OCRVersion.PPOCRV4,
        "Rec.engine_type": EngineType.ONNXRUNTIME,
        "Rec.lang_type": LangRec.CH,
        "Rec.model_type": ModelType.MOBILE,
        "Rec.ocr_version": OCRVersion.PPOCRV4,
        "Cls.engine_type": EngineType.ONNXRUNTIME,
        "Cls.lang_type": LangDet.CH,
        "Cls.model_type": ModelType.MOBILE,
        "Cls.ocr_version": OCRVersion.PPOCRV4,
    }
)

img_url = "https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/test_files/ch_en_num.jpg"
result = engine(img_url)
print(result)

result.vis("vis_result.jpg")

Corresponding configuration file: config.yaml

Configuration File Field Mapping

The following tables are based on default_models.yaml of rapidocr>=3.10.0. engine_type lists only the inference engines that directly register model files in the configuration file; TensorRT dynamically builds the Engine using the corresponding ONNX model. For usage limitations, see the TensorRT inference engine description.

The tiny, small and medium models of PP-OCRv6 are all multilingual models. Under the same specification, different lang_type values route to the same model file:

  • small and medium support: ch, chinese_cht, en, japan, af, az, bs, ca, cs, cy, da, de, es, et, eu, fi, fr, ga, gl, hr, hu, id, is, it, ku, la, lb, lt, lv, mi, ms, mt, nl, no, oc, pl, pt, qu, rm, ro, rs_latin, sk, sl, sq, sv, sw, tl, tr, uz, vi, french, german.
  • tiny supports all of the above languages except japan.
  • Compatible aliases: zh, zh_cn, zh-cn → ch; zh_tw, zh-tw → chinese_cht; ja, jp → japan.

Model Route Details

The model_key in the table below is the actual key in default_models.yaml. lang_type is the route entry; the multi entry of PP-OCRv6 actually accepts the languages listed above, while the multi entry of PP-OCRv5 detection and classification accepts ch, multi.

Module ocr_version lang_type model_type model_key engine_type
det PP-OCRv6 multi tiny multi_PP-OCRv6_det_tiny onnxruntime, openvino, mnn, paddle, torch
det PP-OCRv6 multi small multi_PP-OCRv6_det_small onnxruntime, openvino, mnn, paddle, torch
det PP-OCRv6 multi medium multi_PP-OCRv6_det_medium onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv6 multi tiny multi_PP-OCRv6_rec_tiny onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv6 multi small multi_PP-OCRv6_rec_small onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv6 multi medium multi_PP-OCRv6_rec_medium onnxruntime, openvino, mnn, paddle, torch
det PP-OCRv5 multi mobile ch_PP-OCRv5_det_mobile onnxruntime, openvino, mnn, paddle, torch
det PP-OCRv5 multi server ch_PP-OCRv5_det_server onnxruntime, openvino, mnn, paddle, torch
cls PP-OCRv5 multi mobile ch_PP-LCNet_x0_25_textline_ori_cls_mobile onnxruntime, openvino, mnn, paddle
cls PP-OCRv5 multi server ch_PP-LCNet_x1_0_textline_ori_cls_server onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 ch mobile ch_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv5 korean mobile korean_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 latin mobile latin_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 eslav mobile eslav_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 en mobile en_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 th mobile th_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 el mobile el_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 arabic mobile arabic_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 cyrillic mobile cyrillic_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 devanagari mobile devanagari_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 ta mobile ta_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 te mobile te_PP-OCRv5_rec_mobile onnxruntime, openvino, mnn, paddle
rec PP-OCRv5 ch server ch_PP-OCRv5_rec_server onnxruntime, openvino, mnn, paddle, torch
det PP-OCRv4 ch mobile ch_PP-OCRv4_det_mobile onnxruntime, openvino, mnn, paddle, torch
det PP-OCRv4 en mobile en_PP-OCRv3_det_mobile onnxruntime, openvino, mnn, paddle, torch
det PP-OCRv4 multi mobile multi_PP-OCRv3_det_mobile onnxruntime, openvino, mnn, paddle, torch
det PP-OCRv4 ch server ch_PP-OCRv4_det_server onnxruntime, openvino, mnn, paddle, torch
cls PP-OCRv4 multi mobile ch_ppocr_mobile_v2.0_cls_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 arabic mobile arabic_PP-OCRv4_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 ch mobile ch_PP-OCRv4_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 chinese_cht mobile chinese_cht_PP-OCRv3_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 cyrillic mobile cyrillic_PP-OCRv3_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 devanagari mobile devanagari_PP-OCRv4_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 en mobile en_PP-OCRv4_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 japan mobile japan_PP-OCRv4_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 ka mobile ka_PP-OCRv4_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 korean mobile korean_PP-OCRv4_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 latin mobile latin_PP-OCRv3_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 ta mobile ta_PP-OCRv4_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 te mobile te_PP-OCRv4_rec_mobile onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 ch server ch_PP-OCRv4_rec_server onnxruntime, openvino, mnn, paddle, torch
rec PP-OCRv4 ch_doc server ch_doc_PP-OCRv4_rec_server onnxruntime, openvino, mnn, paddle

Text Detection Models

ocr_version Language type lang_type model_type engine_type
PP-OCRv6 Multilingual the PP-OCRv6 supported languages listed above tiny
small
medium		 onnxruntime
openvino
mnn
paddle
torch
PP-OCRv5 Multilingual ch
multi		 mobile
server		 onnxruntime
openvino
mnn
paddle
torch
PP-OCRv4 Chinese and English ch mobile
server		 onnxruntime
openvino
mnn
paddle
torch
PP-OCRv4 English, Latin en mobile onnxruntime
openvino
mnn
paddle
torch
PP-OCRv4 Multilingual multi mobile onnxruntime
openvino
mnn
paddle
torch

Corresponding usage:

Note

The lang_type field corresponds to LangDet under the Det module.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
from rapidocr import EngineType, LangDet, ModelType, OCRVersion, RapidOCR

engine = RapidOCR(
    params={
        "Det.engine_type": EngineType.TORCH,
        "Det.lang_type": LangDet.CH,
        "Det.model_type": ModelType.MOBILE,
        "Det.ocr_version": OCRVersion.PPOCRV5
    }
)

Text Line Orientation Classification Models

Note

PP-OCRv5 orientation classification models are supported starting from rapidocr>=3.8.0. PP-OCRv4 orientation classification models are still retained in the current configuration.

ocr_version lang_type model_type engine_type
PP-OCRv5 ch
multi		 mobile
server		 onnxruntime
openvino
mnn
paddle
PP-OCRv4 ch
multi		 mobile onnxruntime
openvino
mnn
paddle
torch

Text Recognition Models

Note

The lang_type field corresponds to LangRec under the Rec module.

ocr_version Language type lang_type model_type engine_type
PP-OCRv6 Multilingual the PP-OCRv6 supported languages listed above tiny
small
medium		 onnxruntime
openvino
mnn
paddle
torch
PP-OCRv5 Mixed Chinese, English and Japanese1 ch mobile
server		 onnxruntime
openvino
mnn
paddle
torch
PP-OCRv5 Korean, mixed Latin languages, Slavic, English, Thai, Greek, Arabic, Cyrillic, Devanagari, Tamil, Telugu korean
latin
eslav
en
th
el
arabic
cyrillic
devanagari
ta
te		 mobile onnxruntime
openvino
mnn
paddle
PP-OCRv4 Chinese ch mobile
server		 onnxruntime
openvino
mnn
paddle
torch
PP-OCRv4 Arabic, Traditional Chinese, Cyrillic, Devanagari, English, Japanese, Georgian, Korean, Latin, Tamil, Telugu arabic
chinese_cht
cyrillic
devanagari
en
japan
ka
korean
latin
ta
te		 mobile onnxruntime
openvino
mnn
paddle
torch
PP-OCRv4 Chinese document ch_doc server onnxruntime
openvino
mnn
paddle

PP-OCRv5 recognition models are compatible with ko → korean, ar → arabic; PP-OCRv4 recognition models are additionally compatible with zh_tw, zh-tw → chinese_cht and ja, jp → japan. All versions are compatible with zh, zh_cn, zh-cn → ch.

Usage

The above models can be specified directly through the fields, and the program will download and use them automatically.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
from rapidocr import EngineType, LangDet, ModelType, OCRVersion, RapidOCR

engine = RapidOCR(
    params={
        "Rec.ocr_version": OCRVersion.PPOCRV5,
        "Rec.engine_type": EngineType.PADDLE,
        "Rec.model_type": ModelType.MOBILE,
    }
)

img_url = "https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/test_files/ch_en_num.jpg"
result = engine(img_url)
print(result)

result.vis("vis_result.jpg")

Language Reference Table

lang Language name Alias
latin Latin -
eslav Slavic -
devanagari Devanagari -
cyrillic Cyrillic -
ch_doc Chinese document -
gl Galician -
lb Luxembourgish -
abq Abkhaz -
af Afrikaans -
ang Old English -
ar Arabic arabic
ava Avar -
az Azerbaijani -
be Belarusian -
bg Bulgarian -
bgc Haryanvi -
bh Bihari -
bho Bhojpuri -
bs Bosnian -
ca Catalan -
ch Simplified Chinese -
che Chechen -
chinese_cht Traditional Chinese -
cs Czech -
cy Welsh -
da Danish -
dar Dargwa -
de German german
el Greek -
en English -
es Spanish -
et Estonian -
fa Persian -
fr French french
ga Irish -
gom Konkani -
hi Hindi -
hr Croatian -
hu Hungarian -
id Indonesian -
inh Ingush -
is Icelandic -
it Italian -
japan Japanese -
ka Georgian -
kbd Kabardian -
korean Korean -
ku Kurdish -
la Latin -
lbe Lak -
lez Lezgi -
lt Lithuanian -
lv Latvian -
mah Magahi -
mai Maithili -
mi Maori -
mn Mongolian -
mr Marathi -
ms Malay -
mt Maltese -
ne Nepali -
new Newari -
nl Dutch -
no Norwegian -
oc Occitan -
pi Pali -
pl Polish -
pt Portuguese -
ro Romanian -
rs_cyrillic Serbian (Cyrillic alphabet) -
rs_latin Serbian (Latin alphabet) -
ru Russian -
sa Sanskrit -
sck Sadri -
sk Slovak -
sl Slovenian -
sq Albanian -
sv Swedish -
sw Swahili -
tab Tabassaran -
ta Tamil -
te Telugu -
th Thai -
tl Tagalog -
tr Turkish -
ug Uyghur -
uk Ukrainian -
ur Urdu -
uz Uzbek -
vi Vietnamese -
qu Quechua -
rm Romansh -
eu Basque -
fi Finnish -

  1. Simplified Chinese, Chinese Pinyin, Traditional Chinese, English, Japanese ↩

Comments