paddleocrserver-powered
A self-contained, multi-language PaddleOCR HTTP server packaged as a single
PyPI install. Flask + Waitress, configurable per-language YAML pipelines, and
two preprocessing modes (fast for screens/UI/printed docs, full for
photographed/curved/rotated documents).
Any client (Java, Python, RF, curl) → HTTP POST → Flask → PaddleOCR → JSON
Install
pip install paddleocrserver-powered
The first request for a given (language, mode) pair downloads the corresponding PaddleOCR model weights into the local PaddleX cache.
Quickstart
paddleocrserver-powered
# → serves http://127.0.0.1:5000 (latin language, fast mode)
Run a quick OCR:
curl -X POST http://127.0.0.1:5000/ocr \
-H 'Content-Type: application/json' \
-d '{"image_path": "/abs/path/to/image.png"}'
You can also run it as a module:
python -m paddleocrserver_powered --port 5000
OCR modes
| Mode | Pipeline flags enabled | Latency (indicative) | Use cases |
|---|---|---|---|
fast |
text detection + recognition only | ~150–400 ms / image | UI screenshots, screen captures, clean printed documents |
full |
+ doc orientation classify + textline orientation + doc unwarping | ~600–1500 ms / image | Photographed docs, curved/rotated pages, low-quality scans |
fast is the default. Pass "preprocess": "full" in the request body to
enable the full pipeline for a single call.
Languages
Built-in language packs (PP-OCRv5 mobile recognition models):
| Code | Recognition model |
|---|---|
latin |
latin_PP-OCRv5_mobile_rec (default) |
japan |
japan_PP-OCRv5_mobile_rec |
korean |
korean_PP-OCRv5_mobile_rec |
chinese_cht |
chinese_cht_PP-OCRv5_mobile_rec |
cyrillic |
cyrillic_PP-OCRv5_mobile_rec |
arabic |
arabic_PP-OCRv5_mobile_rec |
devanagari |
devanagari_PP-OCRv5_mobile_rec |
By default only latin is enabled. Activate other languages by editing
languages.yaml:
enabled:
latin: true
japan: true
korean: false
chinese_cht: false
cyrillic: false
arabic: false
devanagari: false
default_language: latin
Then point the server at it:
paddleocrserver-powered --languages-config /path/to/languages.yaml
# or
OCR_LANGUAGES_CONFIG=/path/to/languages.yaml paddleocrserver-powered
All enabled: true languages are eager-loaded at startup in fast mode,
so the first request never pays a cold-start cost. The corresponding full
engine is created lazily the first time someone requests preprocess: "full".
Adding languages
To add a custom language pack:
- Drop two YAML files next to the packaged ones (or beside your custom
languages.yaml):ocr_config_<code>.yamlandocr_config_<code>_full.yaml. - Set the appropriate
model_nameunderSubModules.TextRecognition(e.g.myscript_PP-OCRv5_mobile_rec). - Add
<code>: trueto yourlanguages.yaml.
The shipped YAMLs are a safe template — copy ocr_config_latin.yaml, change
the model_name, you're done.
API endpoints
POST /ocr
Run OCR on an image.
curl -X POST http://127.0.0.1:5000/ocr \
-H 'Content-Type: application/json' \
-d '{
"image_path": "/abs/path/to/image.png",
"lang": "latin",
"preprocess": "fast"
}'
| Field | Type | Required | Default | Notes |
|---|---|---|---|---|
image_path |
string | yes | — | Absolute path on the server side |
lang |
string | no | default_language |
Must be in the enabled list |
preprocess |
string | no | "fast" |
"fast" or "full" |
If lang is not enabled the server returns 400 with the list of
currently-enabled languages.
POST /ocr_strikethrough
OCR + strikethrough detection in one call. Always uses fast preprocessing.
curl -X POST http://127.0.0.1:5000/ocr_strikethrough \
-H 'Content-Type: application/json' \
-d '{"image_path": "/abs/path/img.png", "text": "cancelled", "lang": "latin"}'
GET /status
curl http://127.0.0.1:5000/status
Returns {status, uptime, loaded_engines}.
GET /languages
curl http://127.0.0.1:5000/languages
Returns {enabled, default_language, supported}.
Configuration
| Env var | Default | Purpose |
|---|---|---|
OCR_HOST |
127.0.0.1 |
Bind address |
OCR_PORT |
5000 |
Bind port |
OCR_LANGUAGES_CONFIG |
(packaged) | Path to a custom languages.yaml |
OCR_RATE_LIMIT_DEFAULT |
100 per minute |
Global rate limit (per IP) |
OCR_RATE_LIMIT_OCR |
30 per minute |
Per-IP limit on /ocr and /ocr_strikethrough |
OCR_RATE_LIMIT_STORAGE |
memory:// |
flask-limiter storage URI (e.g. Redis) |
CLI flags (override env vars): --host, --port, --languages-config.
Advanced: multi-port deployment
PaddleOCR is not thread-safe — a single process runs with threads=1. To
serve multiple languages efficiently, run one process per language on its
own port and route from the client side.
/etc/paddleocr/latin.yaml:
enabled: { latin: true }
default_language: latin
/etc/paddleocr/japan.yaml:
enabled: { japan: true }
default_language: japan
Launch:
OCR_PORT=5000 paddleocrserver-powered --languages-config /etc/paddleocr/latin.yaml &
OCR_PORT=5001 paddleocrserver-powered --languages-config /etc/paddleocr/japan.yaml &
Each instance keeps only its own model in RAM and answers only on its port.
Programmatic usage
from paddleocrserver_powered import spawn_server
proc = spawn_server(port=5050, languages_config="/etc/paddleocr/latin.yaml")
try:
# ... do work, hit http://127.0.0.1:5050/ocr ...
pass
finally:
proc.terminate()
proc.wait(timeout=10)
spawn_server returns a subprocess.Popen running python -m paddleocrserver_powered with the provided arguments.
License
MIT — see LICENSE.
Metadata
Release files for paddleocrserver-powered 0.1.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| paddleocrserver_powered-0.1.1.tar.gz | 18.7 kB | Without provenance | Details for paddleocrserver_powered-0.1.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| paddleocrserver_powered-0.1.1-py3-none-any.whl | Python 3 | none | any | Details for paddleocrserver_powered-0.1.1-py3-none-any.whl |
Total release size: 44.9 kB
Release files / paddleocrserver_powered-0.1.1.tar.gz
| Download URL | paddleocrserver_powered-0.1.1.tar.gz |
|---|---|
| Size | 18.7 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
1258bc634731af7341cecdb9c3c8cadd42f94ea4df0bc81a85912806afbd8384
|
|
BLAKE2b-256 checksum How to use checksums |
e4db07045c34094b79b9bce7b61286b4e912991a4bb99fa124615180e1db58c1
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.13
|
Release files / paddleocrserver_powered-0.1.1-py3-none-any.whl
| Download URL | paddleocrserver_powered-0.1.1-py3-none-any.whl |
|---|---|
| Size | 26.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
0a1be44371b1801b268e80cf2516e0d0419867c79c7e1a24713e941d0b29bc28
|
|
BLAKE2b-256 checksum How to use checksums |
f180b15301c90cc90331bbc0c31db3ba4a13496a7b3521fdb3184872ceabd790
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.13
|
Security for paddleocrserver-powered 0.1.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.