models_list_en.md

June 20, 2026 · View on GitHub

OCR Model List

Note: PyTorch .pth model download link: https://pan.baidu.com/s/1r1DELT8BlgxeOP2RqREJEg passcode: 6clx

PaddleOCR model download link: https://pan.baidu.com/s/1getAprT2l_JqwhjwML0g9g passcode: lmv7

1. Text Detection Models
2. Text Recognition Models
3. Text Direction Classification Models
4. Document Preprocessing Models
5. Document Structure Analysis Models (PP-StructureV3)

1. Text Detection Models

Model	Description	Size	Download
PP-OCRv6_medium_det	Medium model, PPLCNetV4 + RepLKPAN, 50 languages	61M	Training / Converter
PP-OCRv6_small_det	Small model, PPLCNetV4 + RepLKFPN, 50 languages	9.8M	Training / Converter
PP-OCRv6_tiny_det	Tiny model, PPLCNetV4 + RepLKFPN(k=5), 49 languages	1.9M	Training / Converter
PP-OCRv5_server_det	Server model, PPLCNetV4 + RepLKPAN	110M	Training / Converter
PP-OCRv5_mobile_det	Mobile model	4.7M	Training / Converter
ch_PP-OCRv4_det	Ultra-lightweight	4.7M	Training
ch_PP-OCRv4_server_det	High-precision	110M	Training
ch_PP-OCRv3_det	Original lightweight	3.8M	Training

2. Text Recognition Models

Model	Description	Size	Download
PP-OCRv6_medium_rec	Medium model, 50 languages	73M	Training / Converter
PP-OCRv6_small_rec	Small model, 50 languages	20M	Training / Converter
PP-OCRv6_tiny_rec	Tiny model, 49 languages	4.3M	Training / Converter
PP-OCRv5_server_rec	Server model, SVTR_HGNet	99M	Training / Converter
PP-OCRv5_mobile_rec	Mobile model, SVTR_LCNet	15M	Training / Converter
ch_PP-OCRv4_rec	Ultra-lightweight	8.5M	Training
ch_PP-OCRv4_server_rec	High-precision	88M	Training

3. Text Direction Classification Models

Model	Description	Size	Download
ch_ppocr_mobile_v2.0_cls	Original model	1.38M	Training

4. Document Preprocessing Models

Model	Description	Size	Download
PP-LCNet_x1_0_doc_ori	Document orientation (0°/90°/180°/270°)	7M	Training / Converter
PP-LCNet_x0_25_textline_ori	Text line orientation (0°/180°), ultra-lightweight	0.96M	Training / Converter
PP-LCNet_x1_0_textline_ori	Text line orientation (0°/180°)	6.5M	Training / Converter
UVDoc	Document unwarping (CGU-Net), SIGGRAPH Asia 2023	30.3M	Training / Converter

5. Document Structure Analysis Models (PP-StructureV3)

5.1 Layout Detection Models

Model	Description	Params	Download
PP-DocLayout-M	[Recommended] PicoDet, LCNet(scale=2.0)+LCPAN+PicoHeadV2, 23 layout classes	5.8M	Training / Converter
PP-DocLayout-S	Lightweight PicoDet, LCNet(scale=0.75)+LCPAN+PicoHeadV2, 23 layout classes	1.2M	Training / Converter

5.2 Table Structure Recognition Models

Model	Description	Params	Download
SLANeXt_wired	ViT Encoder + GRU Attention Decoder, HTML output	~90M	Training / Converter

5.3 Formula Recognition Models

Model	Description	Params	Download
PP-FormulaNet_plus-M	[Recommended] PPHGNetV2_B6 + MBart Decoder (6 layers), LaTeX output	~250M	Training / Converter
PP-FormulaNet-S	Lightweight, PPHGNetV2_B4 + MBart Decoder (2 layers), LaTeX output	~100M	Training / Converter

5.4 Seal Text Detection Models

Model	Description	Params	Download
PP-OCRv4_mobile_seal_det	Seal text detection, PPLCNetV3+RSEFPN+DBHead	~1.5M	Training / Converter

5.5 Pipeline Usage

python ptstructure/predict_structure.py \
    --image_dir=./doc/table/ \
    --output_dir=./output/ \
    --layout_variant=M \
    --use_formula --use_seal

See PP-StructureV3 Porting Guide for details.