ruvector-scipix

name: ruvector-scipix description: "OCR client for scientific documents - extracts LaTeX, MathML, and structured text from equations, papers, and technical diagrams. Use when parsing mathematical equations from images, extracting formulas from research PDFs, or converting scientific figures to structured data."

@ruvector/scipix

OCR client specialized for scientific documents. Extracts LaTeX, MathML, and structured text from mathematical equations, research papers, and technical diagrams with high accuracy.

Quick Reference

Installation

npx @ruvector/scipix@latest

Quick Start

import { Scipix, toLatex, toMathML } from '@ruvector/scipix';

const scipix = new Scipix({
  model: 'scientific-v2',
  device: 'cpu',
});

// Extract LaTeX from an equation image
const latex = await scipix.toLatex('./equation.png');
console.log(latex); // "E = mc^2"

// Get MathML representation
const mathml = await scipix.toMathML('./integral.png');
console.log(mathml); // <math>...</math>

// Full document OCR with structure
const result = await scipix.recognize('./paper-page.png');
console.log(result.text);       // Full text
console.log(result.equations);  // Detected equations as LaTeX
console.log(result.tables);     // Detected tables as structured data
console.log(result.figures);    // Figure bounding boxes

// Process entire PDF
const pages = await scipix.processPDF('./paper.pdf');
for (const page of pages) {
  console.log(`Page ${page.number}: ${page.equations.length} equations`);
}

Core API

Scipix

Main OCR client.

const scipix = new Scipix(config?: ScipixConfig);

ScipixConfig:

scipix.recognize(input)

Full document recognition with structure extraction.

await scipix.recognize(input: string | Buffer): Promise<RecognitionResult>

RecognitionResult:

Equation:

scipix.toLatex(input)

Extract equations as LaTeX only.

await scipix.toLatex(input: string | Buffer): Promise<string[]>

scipix.toMathML(input)

Extract equations as MathML only.

await scipix.toMathML(input: string | Buffer): Promise<string[]>

scipix.processPDF(path, options?)

Process a multi-page PDF.

await scipix.processPDF(path: string, options?: PDFOptions): Promise<PageResult[]>

PDFOptions:

PageResult:

scipix.batch(inputs)

Batch process multiple images.

await scipix.batch(inputs: string[]): Promise<RecognitionResult[]>

scipix.detectRegions(input)

Detect document regions without full OCR.

await scipix.detectRegions(input: string | Buffer): Promise<Region[]>

Region: { type: 'text' | 'equation' | 'table' | 'figure'; bbox: BoundingBox; confidence: number }

Common Patterns

Research Paper Pipeline

const scipix = new Scipix({ model: 'scientific-v2' });

// Process paper and collect all equations
const pages = await scipix.processPDF('./arxiv-paper.pdf');
const allEquations = pages.flatMap(p => p.equations);

console.log(`Found ${allEquations.length} equations`);
for (const eq of allEquations) {
  console.log(`  ${eq.latex} (confidence: ${eq.confidence.toFixed(2)})`);
}

Handwritten Equation Recognition

const scipix = new Scipix({ model: 'handwriting' });
const latex = await scipix.toLatex('./whiteboard-photo.jpg');
console.log(`Recognized: ${latex.join(', ')}`);

CLI Usage

# Recognize an image
npx @ruvector/scipix recognize ./equation.png

# Extract LaTeX from PDF
npx @ruvector/scipix pdf ./paper.pdf --format latex

# Batch process
npx @ruvector/scipix batch ./images/*.png --output results.json

References

• API Reference
• npm