davanstrien/paddleocr-vl15-final-test

Document Processing using PaddleOCR-VL-1.5 (OCR mode)

This dataset contains OCR results from images in davanstrien/ufo-ColPali using PaddleOCR-VL-1.5, an ultra-compact 0.9B SOTA OCR model.

Processing Details

• —Source Dataset: davanstrien/ufo-ColPali
• —Model: PaddlePaddle/PaddleOCR-VL-1.5
• —Task Mode: ocr - General text extraction to markdown format
• —Number of Samples: 10
• —Processing Time: 8.3 min
• —Processing Date: 2026-01-30 11:55 UTC

Configuration

• —Image Column: image
• —Output Column: paddleocr_1.5_ocr
• —Dataset Split: train
• —Batch Size: 1
• —Smart Resize: Enabled
• —Max Output Tokens: 512
• —Backend: Transformers (batch inference)

Model Information

PaddleOCR-VL-1.5 is a state-of-the-art, resource-efficient model for document parsing:

• —🎯 SOTA Performance - 94.5% on OmniDocBench v1.5
• —🧩 Ultra-compact - Only 0.9B parameters
• —📝 OCR mode - General text extraction
• —📊 Table mode - HTML table recognition
• —📐 Formula mode - LaTeX mathematical notation
• —📈 Chart mode - Structured chart analysis
• —🔍 Spotting mode - Text spotting with localization
• —🔖 Seal mode - Seal and stamp recognition
• —🌍 Multilingual - Support for multiple languages
• —⚡ Fast - Efficient batch inference

Task Modes

• —OCR: Extract text content to markdown format
• —Table Recognition: Extract tables to HTML format
• —Formula Recognition: Extract mathematical formulas to LaTeX
• —Chart Recognition: Analyze and describe charts/diagrams
• —Spotting: Text spotting with location information
• —Seal Recognition: Extract text from seals and stamps

Dataset Structure

The dataset contains all original columns plus:

• —paddleocr_1.5_ocr: The extracted content based on task mode
• —inference_info: JSON list tracking all OCR models applied to this dataset

Usage

from datasets import load_dataset
import json

# Load the dataset
dataset = load_dataset("{output_dataset_id}", split="train")

# Access the extracted content
for example in dataset:
    print(example["paddleocr_1.5_ocr"])
    break

# View all OCR models applied to this dataset
inference_info = json.loads(dataset[0]["inference_info"])
for info in inference_info:
    print(f"Task: {info['task_mode']} - Model: {info['model_id']}")

Reproduction

This dataset was generated using the uv-scripts/ocr PaddleOCR-VL-1.5 script:

uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/paddleocr-vl-1.5.py \
    davanstrien/ufo-ColPali \
    <output-dataset> \
    --task-mode ocr \
    --image-column image \
    --batch-size 1

Performance

• —Model Size: 0.9B parameters
• —Benchmark Score: 94.5% SOTA on OmniDocBench v1.5
• —Processing Speed: ~0.02 images/second
• —Backend: Transformers batch inference

Generated with 🤖 UV Scripts