Chakkhu Khmer OCR

Chakkhu Khmer OCR

Reads Khmer and English text from pictures of a text line or a printed page, returning the text, a confidence score and line boxes. A CPU demo with a free API, for people who digitize Khmer documents.

Made by muyleanging on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This is a Gradio demo that reads Khmer and English text from pictures. It uses the Chakkhu-OCR-Khmer model, which has 18.6M parameters and runs on a CPU. You upload one line of text or a page of printed text, and a line finder cuts a page into lines first. A free API endpoint returns the text with a confidence score and line boxes.

Who it suits: People who need Khmer or mixed Khmer and English text out of printed pictures, and developers who want an OCR endpoint to call.

Industry | Who uses it | How they use it | Example
Archives | Digitization staff | Turn printed Khmer pages into text | Read a scanned page of a printed document
Business | Back-office teams | Pull text from printed forms and notices | Extract a line of text from a photo
Research | Language researchers | Test an OCR model on Khmer samples | Check the confidence on a sample line
Software | Developers | Call the read endpoint from their own code | Get text and line boxes for an image

Skip it if your text is handwritten or in another script, since the page speaks only of printed Khmer and English.

Where to use it:

• A web browser, on the Space's page at Hugging Face
• Windows, macOS or Linux with Python, as a local copy
• Your own code, through the free API endpoint named read

Use case | Who | What they get back
Read one text line | Anyone | The text of the line with a confidence score
Read a printed page | Digitization staff | Text for each line the finder cut out, with line boxes
Call the API | Developers | Text, confidence and line boxes in the reply

How to set it up:

• Nothing at all if you only use the online Space
• Python and pip, for a local copy
• The Python packages listed in requirements.txt

1. Open the Space's page at Hugging Face to use it online.
2. For a local copy, install the Hugging Face hub package and download the Space.
3. Move into the new folder and install Gradio at the pinned version.
4. Install the packages from requirements.txt.
5. Start the app with Python.

pip install -U huggingface_hub
hf download muyleanging/Chakkhu-OCR-Demo --repo-type space --local-dir Chakkhu-OCR-Demo
cd Chakkhu-OCR-Demo
pip install gradio==6.31.0
pip install -r requirements.txt
python app.py

The repository holds an examples folder with sample pictures and a folder for the model code. The page does not say what a local run needs beyond the packages.

How to use it: Online, open the Space and upload a picture of one text line or a printed page. The app returns the recognized text. Developers can call the free read endpoint, which returns text, confidence and line boxes.

Tool | What it does
Upload | Takes a picture of one text line or a printed page
Line finder | Cuts a page picture into separate lines
read | The API endpoint that returns text, confidence and line boxes
python app.py | Starts the Gradio app on your computer

Example: You upload a picture of a single printed Khmer line from the examples folder. The app returns the recognized text and how confident it is, and you check that against the picture.

For newcomers and experts: If you are new, try one of the bundled example pictures first. They show what a clean printed Khmer line looks like before you use your own scan.

If you know the field, the model is small at 18.6M parameters, and the page reports a character error rate of 4.50 percent on the full KHOB benchmark, against 5.87 percent for Netra-OCR. Those figures are the author's own, so test on your own documents.

You upload a line or a page
The line finder splits a page into lines
The OCR model reads each line
Text, confidence and boxes come back

Good to know:

• License: apache-2
• Cost to run: none named, and the Space sits on the basic CPU tier at Hugging Face
• Data it reaches: the pictures you upload, and the model files from Hugging Face
• Not for: handwriting or other scripts, since the page describes printed Khmer and English

Title: Character error rate on the KHOB benchmark
Chakkhu OCR: 4.50 percent
Netra-OCR: 5.87 percent

On Hugging Face, 13 November 2026

• Likes: 2
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/muyleanging/Chakkhu-OCR-Demo
• Files: https://huggingface.co/spaces/muyleanging/Chakkhu-OCR-Demo/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.