A high-performance highly-customizable reverse OCR tool that renders text or huggingface-compatible datasets to images. Dimension, DPI, CSS configurable!
Reason this release was yanked:
missing dep linkify-it-py
Project description
DeOCR
DeOCR (de-cor), A reverse OCR tool that renders huggingface-compatible datasets to configurable images (e.g., custom size 512x512, black background, paddings, margins, etc.). This tool can be considered as a text-to-image data pre-processing component in pipelines such as DeepSeek-OCR.
---
title: DeOCR Usage in LLM Pipeline
---
flowchart LR
TEXTDATA[/"context as pure text"/]
MMDATA[/"Does this particular car <br/> <image> present in here <image> ?"/]
HFDATASET[("huggingface dataset")]
subgraph DeOCR
CSS1["cli --style red-text,bold"]
CSS2["cli --style default"]
CSS3["cli --style default"]
MAPPER["DeOCR Dataset Mapper"]
end
TEXTDATA --> CSS1 --> IMG1[["🖼️🖼️🖼️🖼️🖼️🖼️🖼️🖼️<br/>🖼️ context as img 🖼️<br/>🖼️🖼️🖼️🖼️🖼️🖼️🖼️🖼️<br/>"]]:::redText
TEXTDATA --> CSS2 --> IMG2[["🖼️🖼️🖼️🖼️🖼️🖼️🖼️🖼️<br/>🖼️ context as img 🖼️<br/>🖼️🖼️🖼️🖼️🖼️🖼️🖼️🖼️<br/>"]]
MMDATA --> CSS3 --> IMG3[["Does this particular car <br/> 🖼️🖼️🖼️🖼️🖼️🖼️🖼️<br/>🖼️🖼️🖼️🚗🖼️🖼️🖼️<br/>🖼️🖼️🖼️🖼️🖼️🖼️🖼️<br/> present in here <br/> 🖼️🖼️🖼️🖼️🖼️🖼️🖼️<br/>🖼️🖼️🖼️🖼️🖼️🖼️🖼️<br/>🖼️🖼️🖼️🖼️🖼️🖼️🖼️<br/>?"]]
HFDATASET --> MAPPER --> DEOCRDATASET[("🖼️ imagified dataset")]
DEOCRDATASET & IMG1 & IMG2 & IMG3 -.-> MODEL["LLMs or VLMs<br/> Evaluation"]
classDef redText color:#ff0000,font-weight:bold;
IMG1 ~~~|"fa:fa-mobile-screen A screenshot of text <br/>w. special formatting"| IMG1
IMG2 ~~~|"fa:fa-mobile-screen A plain screenshot of text"| IMG2
IMG3 ~~~|"fa:fa-mobile-screen A screenshot of both text and images"| IMG3
Here is an output example, sized `512x512`, with random string as context
Quick Start
pip install deocr[playwright,pymupdf]
# activate your python environment, then install playwright deps
playwright install chromium
Alternatively, install from source
# uv
uv add "deocr[playwright,pymupdf] @ git+https://github.com/Moenupa/DeOCR.git"
# activate your python environment, then install playwright deps
playwright install chromium
For development
Please use uv to manage the environment:
git clone https://github.com/Moenupa/DeOCR.git
cd DeOCR
uv venv
uv sync --all-extras --all-groups
source .venv/bin/activate
playwright install chromium
pre-commit install
Known Issues
- async function timeout: increase threshold 0.05 at datasets/utils/py_utils.py:612-626
Performance
DeOCR is mainly optimized by asynchronous rendering and multiprocessing dataset mapping. The rendering speed may vary depending on the machine configuration and the complexity of the text to be rendered. On a standard machine with 32 cores, DeOCR can render more than 1k images per second.
GSM8K dataset (one 512x512 image per sample) rendering speed with Intel Xeon Gold 6430:
# increase MAX_ASYNC_PAGES for more cores
$ MAX_ASYNC_PAGES=1 python tests/dataset/manual_load.py
Map (num_proc=1): 100%|██████████████| 7473/7473 [02:48<00:00, 44.33 examples/s]
Map (num_proc=1): 100%|██████████████| 1319/1319 [00:27<00:00, 47.28 examples/s]
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file deocr-0.3.1.tar.gz.
File metadata
- Download URL: deocr-0.3.1.tar.gz
- Upload date:
- Size: 10.1 kB
- Tags: Source
- Uploaded using Trusted Publishing? Yes
- Uploaded via: uv/0.9.18 {"installer":{"name":"uv","version":"0.9.18","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
3284ee058c93ba5049065afd819c7c9e2c6b7ecae5a19b38dd033aa9680ae6db
|
|
| MD5 |
235fc340cce705eebc15f60c85d25ab1
|
|
| BLAKE2b-256 |
e19d4193890d459201ab3481ed77cd8bb3018fe33bc17e4ce85799d6c947dfce
|
File details
Details for the file deocr-0.3.1-py3-none-any.whl.
File metadata
- Download URL: deocr-0.3.1-py3-none-any.whl
- Upload date:
- Size: 14.8 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? Yes
- Uploaded via: uv/0.9.18 {"installer":{"name":"uv","version":"0.9.18","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
aeae5cf0d8467b0ed78618e16f8b3641db64b8702272cb7c9ce4454e341ef97a
|
|
| MD5 |
fb029a6c85e903f4f55e7b66b24a485e
|
|
| BLAKE2b-256 |
1e40bb31d1b251d399c527c0d57f03c2de23dc65a75ee0efcbab95fc7b82623c
|