Simple-GLM-OCR

Simple optical character recognition based on GLM-OCR with fewer dependencies.

Example

from simpleglmocr import SimpleGlmOcr

model = SimpleGlmOcr()

text = model.run("Text Recognition:", "testimage.jpg")

print(text)

This will print the following text for the image shown below:

Hello, GLM-OCR!
This is a test image.
The quick brown fox jumps
over the lazy dog.

Image

Installation

• Prerequisites: A Python environment with python, pip and git.
• First, install PyTorch according to the instructions on their website.
• Next, install the following Python libraries with pip:

pip install regex numpy pillow safetensors

• Now you can clone the repository and run the example.

git clone https://github.com/99991/Simple-GLM-OCR.git
cd Simple-GLM-OCR
python example.py

Server

You can start a server for a web-based OCR experience by running the following command in the Simple-GLM-OCR directory:

python server.py

You can then visit the website at http://127.0.0.1:8000 to upload images for text recognition, or you can use the API (see below).

API

After you have started the server, you can use the API (requires pip install requests):

import requests

url = "http://127.0.0.1:8000/api/ocr"

# We thank Obama for providing his photo for testing purposes
filename = "obama.jpg"

prompt = """
{
    "last_name": "",
    "first_name": "",
    "tie color": "",
    "facial expression": "",
    "age": "",
    "body posture": "",
    "background": "",
}
"""

with open(filename, "rb") as f:
    image_bytes = f.read()

files = {'image': (filename, image_bytes, 'image/jpeg')}

response = requests.post(url, files=files, data={'prompt': prompt})
response.raise_for_status()

print(response.text)

Image

Output

```json
{
    "last_name": "OBAMA",
    "first_name": "BARACK",
    "tie color": "blue",
    "facial expression": "smiling",
    "age": "47",
    "body posture": "crossed arms",
    "background": "American flag and presidential seal"
}
```

cURL

You can also use cURL to send a text recognition request to the server:

curl -X POST \
    -F 'prompt=Text Recognition:' \
    -F 'image=@testimage.jpg' \
    http://127.0.0.1:8000/api/ocr

This makes it very easy to build a screen text recognition tool using the scrot program.

#!/usr/bin/env bash
scrot -s -o /tmp/capture.png
curl -X POST \
    -F 'prompt=Text Recognition:' \
    -F 'image=@/tmp/capture.png' \
    http://127.0.0.1:8000/api/ocr > /tmp/text.txt
xdg-open /tmp/text.txt

Put the code in a file, mark it as executable and bind it to a shortcut for convenient access!

Prompt Formats

GLM-OCR supports multiple prompt formats:

• Text Recognition: (for general text recognition)
• Table Recognition: (for tables as HTML)
• Formula Recognition: (for equations in LaTeX)
• Schema-based JSON extraction

FAQ

• How to run without GPU?
  • Load the model in CPU-mode: model = SimpleGlmOcr(device="cpu")