Skip to content

models.md

Latest commit

 

History

History
246 lines (195 loc) · 25.2 KB

models.md

File metadata and controls

246 lines (195 loc) · 25.2 KB
Raw

Pre-trained Models

OAROCR provides pre-trained models for OCR and document understanding tasks. Download them manually from the GitHub Releases page (linked in the tables below), or have the library fetch them on demand from ModelScope. See Auto-download at the bottom.

Text Detection Models

Choose between mobile and server variants based on your needs:

  • Mobile: Smaller, faster models suitable for real-time applications
  • Server: Larger, more accurate models for high-precision requirements
Version Category Model File Size Description
PP-OCRv4 Mobile pp-ocrv4_mobile_det.onnx 4.6 MiB Mobile variant for real-time applications
PP-OCRv4 Server pp-ocrv4_server_det.onnx 108.2 MiB Server variant for high-precision
PP-OCRv5 Mobile pp-ocrv5_mobile_det.onnx 4.6 MiB Mobile variant for real-time applications
PP-OCRv5 Server pp-ocrv5_server_det.onnx 84.0 MiB Server variant for high-precision

Text Recognition Models

Chinese/General Models

Version Category Model File Size Description
PP-OCRv3 Mobile pp-ocrv3_mobile_rec.onnx 10.2 MiB Legacy mobile variant
PP-OCRv4 Mobile pp-ocrv4_mobile_rec.onnx 10.4 MiB Mobile variant
PP-OCRv4 Server pp-ocrv4_server_rec.onnx 86.3 MiB Server variant
PP-OCRv4 Document pp-ocrv4_server_rec_doc.onnx 90.5 MiB Optimized for documents
PP-OCRv5 Mobile pp-ocrv5_mobile_rec.onnx 15.8 MiB Mobile variant
PP-OCRv5 Server pp-ocrv5_server_rec.onnx 80.6 MiB Server variant
SVTRv2 Server ch_svtrv2_rec.onnx 80.3 MiB High accuracy variant
RepSVTR Server ch_repsvtr_rec.onnx 24.2 MiB Balanced accuracy/speed

Language-Specific Models

Version Language Model File Size Description
PP-OCRv3 Arabic arabic_pp-ocrv3_mobile_rec.onnx 8.6 MiB Arabic text recognition
PP-OCRv5 Arabic arabic_pp-ocrv5_mobile_rec.onnx 7.7 MiB Arabic text recognition
PP-OCRv3 Chinese Traditional chinese_cht_pp-ocrv3_mobile_rec.onnx 10.6 MiB Traditional Chinese text recognition
PP-OCRv3 Cyrillic cyrillic_pp-ocrv3_mobile_rec.onnx 8.6 MiB Cyrillic script recognition
PP-OCRv5 Cyrillic cyrillic_pp-ocrv5_mobile_rec.onnx 7.7 MiB Cyrillic script recognition
PP-OCRv3 Devanagari devanagari_pp-ocrv3_mobile_rec.onnx 8.6 MiB Devanagari script recognition
PP-OCRv5 Devanagari devanagari_pp-ocrv5_mobile_rec.onnx 7.6 MiB Devanagari script recognition
PP-OCRv5 Greek el_pp-ocrv5_mobile_rec.onnx 7.5 MiB Greek text recognition
PP-OCRv3 English en_pp-ocrv3_mobile_rec.onnx 8.6 MiB English text recognition
PP-OCRv4 English en_pp-ocrv4_mobile_rec.onnx 7.4 MiB English text recognition
PP-OCRv5 English en_pp-ocrv5_mobile_rec.onnx 7.5 MiB English text recognition
PP-OCRv5 Eastern Slavic eslav_pp-ocrv5_mobile_rec.onnx 7.5 MiB Eastern Slavic languages
PP-OCRv3 Japanese japan_pp-ocrv3_mobile_rec.onnx 9.6 MiB Japanese text recognition
PP-OCRv3 Georgian ka_pp-ocrv3_mobile_rec.onnx 8.6 MiB Georgian text recognition
PP-OCRv3 Korean korean_pp-ocrv3_mobile_rec.onnx 9.5 MiB Korean text recognition
PP-OCRv5 Korean korean_pp-ocrv5_mobile_rec.onnx 12.8 MiB Korean text recognition
PP-OCRv3 Latin latin_pp-ocrv3_mobile_rec.onnx 8.6 MiB Latin script recognition
PP-OCRv5 Latin latin_pp-ocrv5_mobile_rec.onnx 7.7 MiB Latin script recognition
PP-OCRv3 Tamil ta_pp-ocrv3_mobile_rec.onnx 8.6 MiB Tamil text recognition
PP-OCRv5 Tamil ta_pp-ocrv5_mobile_rec.onnx 7.5 MiB Tamil text recognition
PP-OCRv3 Telugu te_pp-ocrv3_mobile_rec.onnx 8.6 MiB Telugu text recognition
PP-OCRv5 Telugu te_pp-ocrv5_mobile_rec.onnx 7.6 MiB Telugu text recognition
PP-OCRv5 Thai th_pp-ocrv5_mobile_rec.onnx 7.6 MiB Thai text recognition

PP-OCRv6

PP-OCRv6 is the newest PP-OCR generation. The flat ONNX files and dictionaries are published in this project's v0.7.0 release and on ModelScope. The tables also link the original PaddlePaddle inference bundles.

Source / attribution. Published by PaddlePaddle under the PaddleOCR project (Apache-2.0).

Detection

Size Auto-download ONNX File size Official bundle
tiny pp-ocrv6_tiny_det.onnx 1.7 MiB PP-OCRv6_tiny_det_onnx_infer.tar
small pp-ocrv6_small_det.onnx 9.4 MiB PP-OCRv6_small_det_onnx_infer.tar
medium pp-ocrv6_medium_det.onnx 59.2 MiB PP-OCRv6_medium_det_onnx_infer.tar

Recognition

Size Auto-download ONNX Dictionary File size Official bundle
tiny pp-ocrv6_tiny_rec.onnx ppocrv6_tiny_dict.txt (6904 characters) 4.3 MiB PP-OCRv6_tiny_rec_onnx_infer.tar
small pp-ocrv6_small_rec.onnx ppocrv6_dict.txt (18708 characters) 20.2 MiB PP-OCRv6_small_rec_onnx_infer.tar
medium pp-ocrv6_medium_rec.onnx ppocrv6_dict.txt (18708 characters) 73.0 MiB PP-OCRv6_medium_rec_onnx_infer.tar

Character Dictionaries

Character dictionaries are required for text recognition. Choose the appropriate dictionary for your model:

General Dictionaries

Version File Description
PP-OCRv4 Document ppocrv4_doc_dict.txt For PP-OCRv4 document models
PP-OCRv5 ppocrv5_dict.txt For PP-OCRv5 models
PP-OCRv6 Tiny ppocrv6_tiny_dict.txt For PP-OCRv6 tiny recognition
PP-OCRv6 Small/Medium ppocrv6_dict.txt For PP-OCRv6 small and medium recognition
PP-OCR Keys v1 ppocr_keys_v1.txt For PP-OCRv3 general and PP-OCRv4 general models

PP-OCRv5 Language-Specific Dictionaries

Language File Model Compatibility
Arabic ppocrv5_arabic_dict.txt PP-OCRv5 Arabic
Cyrillic ppocrv5_cyrillic_dict.txt PP-OCRv5 Cyrillic
Devanagari ppocrv5_devanagari_dict.txt PP-OCRv5 Devanagari
Greek ppocrv5_el_dict.txt PP-OCRv5 Greek
English ppocrv5_en_dict.txt PP-OCRv5 English
Eastern Slavic ppocrv5_eslav_dict.txt PP-OCRv5 Eastern Slavic
Korean ppocrv5_korean_dict.txt PP-OCRv5 Korean
Latin ppocrv5_latin_dict.txt PP-OCRv5 Latin script
Tamil ppocrv5_ta_dict.txt PP-OCRv5 Tamil
Telugu ppocrv5_te_dict.txt PP-OCRv5 Telugu
Thai ppocrv5_th_dict.txt PP-OCRv5 Thai

PP-OCRv3 and PP-OCRv4 Language-Specific Dictionaries

The language-specific PP-OCRv3 dictionaries are different from the PP-OCRv5 dictionaries and are not part of the auto-download registry. Download them from PaddleOCR and pass the local path to the recognition builder. The PP-OCRv4 English checkpoint uses the same en_dict.txt file as PP-OCRv3 English.

Language Official dictionary Model compatibility
Arabic arabic_dict.txt PP-OCRv3 Arabic
Traditional Chinese chinese_cht_dict.txt PP-OCRv3 Traditional Chinese
Cyrillic cyrillic_dict.txt PP-OCRv3 Cyrillic
Devanagari devanagari_dict.txt PP-OCRv3 Devanagari
English en_dict.txt PP-OCRv3 and PP-OCRv4 English
Japanese japan_dict.txt PP-OCRv3 Japanese
Georgian ka_dict.txt PP-OCRv3 Georgian
Korean korean_dict.txt PP-OCRv3 Korean
Latin latin_dict.txt PP-OCRv3 Latin script
Tamil ta_dict.txt PP-OCRv3 Tamil
Telugu te_dict.txt PP-OCRv3 Telugu

Preprocessing Models

Models for document preprocessing and orientation detection:

Type Model File Size Description
Document Orientation pp-lcnet_x1_0_doc_ori.onnx 6.5 MiB Detect document rotation
Text Line Orientation (Light) pp-lcnet_x0_25_textline_ori.onnx 995 KiB Fast text line orientation
Text Line Orientation pp-lcnet_x1_0_textline_ori.onnx 6.5 MiB Accurate text line orientation
Document Rectification uvdoc.onnx 30.2 MiB Fix perspective distortion

Document Structure Models

Models for document structure analysis with OARStructureBuilder:

Layout Detection

Model Model File Size Description
PicoDet-L 17cls picodet-l_layout_17cls.onnx 22.4 MiB 17-class layout detection
PicoDet-L 3cls picodet-l_layout_3cls.onnx 22.4 MiB 3-class layout detection
PicoDet-S 17cls picodet-s_layout_17cls.onnx 4.7 MiB Fast 17-class layout
PicoDet-S 3cls picodet-s_layout_3cls.onnx 4.7 MiB Fast 3-class layout
PicoDet 1x picodet_layout_1x.onnx 7.2 MiB Legacy layout model
PicoDet 1x Table picodet_layout_1x_table.onnx 7.2 MiB Table-focused layout
PP-DocLayout-S pp-doclayout-s.onnx 4.7 MiB Small variant
PP-DocLayout-M pp-doclayout-m.onnx 22.4 MiB Medium variant
PP-DocLayout-L pp-doclayout-l.onnx 123.4 MiB Large variant
PP-DocLayout_plus-L pp-doclayout_plus-l.onnx 123.7 MiB Enhanced large variant
PP-DocLayoutV2 pp-doclayoutv2.onnx 204.1 MiB V2 with reading order (col, row)
PP-DocLayoutV3 pp-doclayoutv3.onnx 123.9 MiB V3 with single order key
PP-DocBlockLayout pp-docblocklayout.onnx 123.3 MiB Hierarchical ordering
RT-DETR-H 17cls rt-detr-h_layout_17cls.onnx 469.3 MiB High accuracy 17-class
RT-DETR-H 3cls rt-detr-h_layout_3cls.onnx 469.2 MiB High accuracy 3-class

Table Recognition

Component Model File Size Description
Table Classification pp-lcnet_x1_0_table_cls.onnx 6.5 MiB Wired vs wireless
Cell Detection (Wired) rt-detr-l_wired_table_cell_det.onnx 123.3 MiB RT-DETR for wired tables
Cell Detection (Wireless) rt-detr-l_wireless_table_cell_det.onnx 123.3 MiB RT-DETR for wireless tables
Structure (SLANet) slanet.onnx 7.4 MiB Basic structure recognition
Structure (SLANet+) slanet_plus.onnx 7.4 MiB Wireless table structure
Structure (SLANeXt Wired) slanext_wired.onnx 350.7 MiB High accuracy wired structure
Structure (SLANeXt Wireless) slanext_wireless.onnx 350.7 MiB High accuracy wireless structure
Structure Dictionary table_structure_dict_ch.txt - Required for structure recognition

Formula Recognition

Model Model File Size Description
PP-FormulaNet-S pp-formulanet-s.onnx 221.1 MiB Small variant
PP-FormulaNet-L pp-formulanet-l.onnx 696.5 MiB Large variant
PP-FormulaNet_plus-S pp-formulanet_plus-s.onnx 221.1 MiB Enhanced small variant
PP-FormulaNet_plus-M pp-formulanet_plus-m.onnx 564.9 MiB Enhanced medium variant
PP-FormulaNet_plus-L pp-formulanet_plus-l.onnx 699.5 MiB Enhanced large variant
PP-FormulaNet Tokenizer pp-formulanet-tokenizer.json 2.0 MiB Required for PP-FormulaNet variants
UniMERNet unimernet.onnx 1.7 GiB Unified Math Expression Recognition
UniMERNet Tokenizer unimernet_tokenizer.json 2.0 MiB Required for UniMERNet

Seal Text Detection

Model Model File Size Description
Seal Detection (Mobile) pp-ocrv4_mobile_seal_det.onnx 4.6 MiB Fast seal detection
Seal Detection (Server) pp-ocrv4_server_seal_det.onnx 108.2 MiB Accurate seal detection

Auto-download

cargo add oar-ocr --features auto-download
use oar_ocr::prelude::*;
let ocr = OAROCRBuilder::new(
    "pp-ocrv5_mobile_det.onnx",   // bare name resolved through the registry
    "pp-ocrv5_mobile_rec.onnx",
    "ppocrv5_dict.txt",
).build()?;
# Ok::<(), Box<dyn std::error::Error>>(())

When the feature is enabled, registered file names are fetched from greatv/oar-ocr on ModelScope into $OAR_HOME (default ~/.oar) and verified against the expected SHA-256 before use. Subsequent runs reuse the cached copy. The bundled registry lives at oar_ocr::download::REGISTRY.

Path resolution rules

These rules apply only to path sources. Models passed as in-memory bytes (see Loading Models from Memory) bypass path resolution entirely.

For each model path argument the builder applies these checks in order:

  1. Existing file wins. If the path refers to a real file on disk it is used as-is — no registry lookup, no hash check, no network. A ./pp-ocrv5_mobile_det.onnx next to the binary always shadows the registry.
  2. Only bare names or $OAR_HOME-rooted paths are eligible for auto-download. A path is considered for registry resolution only when it has no parent component (e.g. "pp-ocrv5_mobile_det.onnx") or when its parent equals the cache directory. Explicit paths with another parent component are returned verbatim even if their file name is registered — the library never silently overrides an explicit path.
  3. Registry hit: cache or download. If the file name appears in REGISTRY:
    • The cached copy is used without network access when $OAR_HOME/<name> exists with the expected size and SHA-256.
    • Otherwise, the file is downloaded from ModelScope, verified with SHA-256, and atomically replaced.
  4. Unregistered and missing. The path is returned verbatim so the builder produces its normal "model not found" error.
Input On disk Behaviour
"pp-ocrv5_mobile_det.onnx" ./pp-ocrv5_mobile_det.onnx exists Use the local CWD file
"pp-ocrv5_mobile_det.onnx" $OAR_HOME/... exists, hash OK Use cached copy, no network
"pp-ocrv5_mobile_det.onnx" absent or hash mismatch Download to $OAR_HOME, verify, use
"custom/det.onnx" absent Returned as-is, resulting in "model not found"
"$OAR_HOME/pp-ocrv5_mobile_det.onnx" (absolute) (any) Handled like a bare name because its parent is the cache directory

Note: the resolver compares paths verbatim — ~ is not expanded. Pass a bare filename, an absolute path under $OAR_HOME, or let your shell expand ~ for you.

Cache layout

  • Override the cache root with the OAR_HOME environment variable. Defaults to ~/.oar (resolved via the platform home directory, while the literal ~ is not expanded by the library).
  • Files land at $OAR_HOME/<name>, flat (no per-revision subdirectories).
  • Downloads stream into a unique $OAR_HOME/.<name>.<pid>.<n>.part and are renamed atomically once the SHA-256 matches, so a crash mid-download won't poison the cache and concurrent processes don't clobber each other.
  • After verification a $OAR_HOME/.<name>.sha256 sidecar records the verified hash. Future loads with a matching cache file + sidecar skip the multi-second rehash. Deleting the sidecar forces a fresh hash check.