README.md

metadata

title: Bec Dot.orc Api
emoji: 🚀
colorFrom: purple
colorTo: red
sdk: gradio
sdk_version: 6.5.1
app_file: app.py
pinned: false
license: apache-2.0

Bec Dot.ocr API

OCR API powered by rednote-hilab/dots.ocr -- a multilingual document-parsing vision-language model. This Space provides both a browser UI and a programmatic API optimized for batch processing.

Quick start

1. Install the client

pip install gradio_client

2. Process a single image

from gradio_client import Client

client = Client("openpecha/bec-dot.orc-api")

result = client.predict(
    "path/to/image.png",                        # local filepath or URL
    "Extract the text content from this image.", # prompt
    api_name="/predict",
)
print(result)

3. Batch-process many images

import os
import json
from pathlib import Path
from gradio_client import Client, handle_file

client = Client("openpecha/bec-dot.orc-api")

image_dir = Path("images")
output_dir = Path("results")
output_dir.mkdir(exist_ok=True)

prompt = "Extract the text content from this image."

for img_path in sorted(image_dir.glob("*.png")):
    print(f"Processing {img_path.name} ...")
    result = client.predict(
        handle_file(str(img_path)),
        prompt,
        api_name="/predict",
    )
    out_file = output_dir / f"{img_path.stem}.txt"
    out_file.write_text(result, encoding="utf-8")
    print(f"  -> saved to {out_file}")

Tip: The Space uses queuing (max_size=20), so requests are processed sequentially and will not time out even for large batches.

4. Use a custom prompt

The default prompt is "Extract the text content from this image." You can override it for more specific tasks:

# Layout-aware JSON extraction
result = client.predict(
    handle_file("document.png"),
    """Please output the layout information from the PDF image, including each layout element's bbox, its category, and the corresponding text content within the bbox.

1. Bbox format: [x1, y1, x2, y2]
2. Layout Categories: ['Caption', 'Footnote', 'Formula', 'List-item', 'Page-footer', 'Page-header', 'Picture', 'Section-header', 'Table', 'Text', 'Title'].
3. Text Extraction & Formatting Rules:
    - Picture: omit the text field.
    - Formula: format as LaTeX.
    - Table: format as HTML.
    - All Others: format as Markdown.
4. Output the original text with no translation.
5. Sort all layout elements in human reading order.
6. Final Output: a single JSON object.""",
    api_name="/predict",
)

API reference

Endpoint	Method	Parameters	Returns
/predict	POST	image (filepath/URL), prompt (string)	Raw text or JSON string

Model details

• Model: rednote-hilab/dots.ocr (1.7B LLM, ~3B total)
• Precision: bfloat16
• Capabilities: text extraction, layout detection, table recognition (HTML), formula parsing (LaTeX), multilingual support