poster2json

Convert scientific posters (PDF/images) to structured JSON metadata using Large Language Models.

Overview

poster2json extracts structured metadata from scientific conference posters into machine-actionable JSON conforming to the poster-json-schema.

Features

  • PDF Processing: Layout-aware text extraction via pdfalto
  • Image Processing: Vision-based OCR via Qwen2-VL-7B
  • JSON Structuring: Fine-tuned Llama 3.1 8B for poster-specific metadata
  • Schema Validation: Built-in validation against poster-json-schema
  • CLI & Python API: Flexible usage options

Quick Start

Installation

pip install poster2json

CLI Usage

# Extract metadata from a poster
poster2json extract poster.pdf -o result.json

# Validate extracted JSON
poster2json validate result.json

Python API

from poster2json import extract_poster, validate_poster

# Extract metadata
result = extract_poster("poster.pdf")
print(result["titles"][0]["title"])

# Validate
is_valid = validate_poster(result)

Architecture

The pipeline processes posters in two stages:

  1. Raw Text Extraction
  2. PDF files → pdfalto (layout-aware XML)

  3. Image files → Qwen2-VL-7B (vision OCR)

  4. JSON Structuring

  5. Raw text → Llama 3.1 8B → Structured JSON

See Architecture for technical details.

Performance

Validated on 10 manually annotated scientific posters with 100% pass rate.

See Evaluation for detailed metrics.

Requirements

  • NVIDIA GPU with ≥16GB VRAM
  • Python 3.10+
  • pdfalto (for PDF processing)