Manual Poster Annotation Ground Truth Dataset

This directory contains manually annotated scientific posters that serve as ground truth for machine-actionable poster metadata extraction.

Purpose

These annotations provide:

• Training data for automated poster extraction models
• Validation benchmarks for AI extraction quality assessment
• Examples of complete, machine-actionable poster metadata

Directory Structure

Each poster is contained in its own folder named by a unique identifier (e.g., 42/, 10890106/). Within each folder, there are four files representing different aspects of the poster annotation:

{poster_id}/
├── {poster_id}.pdf          # Original poster file
├── {poster_id}_raw.md       # Extracted text in structured markdown
├── {poster_id}.json         # Full metadata JSON (complete schema)
├── {poster_id}_sub-json.json # Poster content subset (AI-ready)

---

File Descriptions

1. Original Poster File ({poster_id}.pdf or {poster_id}.jpg)

The source poster file in its original format. This is the visual artifact from which all metadata is extracted.

Accepted File Formats:

Primary Formats (Recommended)

Format	Extension	Notes
PDF	.pdf	Most common poster format. Must be single-page.
PNG	.png	Lossless image format.
JPEG	.jpg, .jpeg	Widely used image format.
TIFF	.tiff, .tif	High-quality, print-ready.
PowerPoint	.pptx	Must be single-slide.
SVG	.svg	Vector format from Figma/Illustrator.

Additional Supported Formats

Format	Extension	Notes
PowerPoint Legacy	.ppt	Older PowerPoint format.
OpenDocument	.odp	LibreOffice/OpenOffice presentations.
Keynote	.key	Apple Keynote (macOS).
EPS	.eps	Encapsulated PostScript vector.
WebP	.webp	Modern web image format.
GIF	.gif	Limited colors, rare for posters.
BMP	.bmp	Uncompressed bitmap.
HEIF	.heic, .heif	Apple high-efficiency format.

Page/Slide Requirements

For multi-page formats (PDF, PPTX, PPT, ODP, KEY), only single-page/single-slide documents are accepted as valid posters. Multi-page documents are automatically flagged for review.

Design Tool Exports

If you created your poster in design software like Figma, Adobe Illustrator, or Canva, we recommend exporting as:

1. PDF (preferred) - maintains quality and is widely compatible
2. SVG - preserves vector quality
3. PNG - high-resolution raster fallback

---

2. Structured Markdown ({poster_id}_raw.md)

Human-readable text extracted from the poster, organized into logical sections. This file preserves the poster's textual content with semantic structure using markdown formatting.

Characteristics:

• Title and author information at the top
• Section headers matching poster sections (Abstract, Introduction, Methods, Results, etc.)
• Preserved figure/table references
• Citation and reference formatting
• Contact information

Example structure:

# Poster Title

Author Name¹, Co-Author Name²

¹Institution 1; ²Institution 2

**Poster ID:** #XXX

## Abstract
[Abstract text...]

## Introduction
[Introduction text...]

## Methodology
[Methods text...]

## Results
[Results text with figure references...]

## Discussion/Conclusions
[Conclusions text...]

## References
[Formatted citations...]

---

3. Full Metadata JSON ({poster_id}.json)

Complete poster metadata including all available information:

• Identifiers: DOIs, poster numbers, file names
• Creators: Author names, ORCIDs, affiliations with ROR identifiers
• Titles: Main and alternative titles
• Publisher: Conference organizer or hosting institution
• Publication Year: Year of presentation
• Subjects: Keywords and classification terms
• Dates: Presentation dates, conference dates
• Language: Primary language (ISO 639)
• Types: Resource type (Conference Poster)
• Related Identifiers: DOIs to papers, datasets, software
• Rights: License information
• Descriptions: Abstract, Methods, Technical Info
• Funding References: Grant numbers, funder information
• Conference: Full conference metadata
• Poster Content: Structured section content
• Image/Table Captions: All figure and table captions
• Domain: Research field classification

---

4. Poster Content Sub-JSON ({poster_id}_sub-json.json)

A subset of the full JSON containing only the poster-derived content. This file is optimized for:

• AI/ML model training and inference
• Lightweight API responses
• Quick content preview

Contents include:

• Basic identifiers
• Minimal creator information (names, contact, affiliations)
• Title
• Poster content sections (Abstract, Introduction, Methods, Results, etc.)
• Image and table captions

Key difference from full JSON:

• Excludes external metadata (ORCIDs, ROR identifiers, DOIs)
• Excludes funding and rights information
• Excludes conference administrative details
• Focuses on what can be extracted directly from the poster image

---

Annotated Posters

ID	Domain	Conference	Year
42	Biomedical Informatics	AACR Annual Meeting	2025
4737132	-	-	-
5128504	-	-	-
6724771	-	-	-
8228476	-	-	-
8228568	-	-	-
10890106	-	-	-
15963941	-	-	-
16083265	-	-	-
17268692	-	-	-

---

Contributing

To add a new manual annotation:

1. Create a folder with a unique identifier: {poster_id}/
2. Add the original poster file: {poster_id}.pdf (or appropriate format)
3. Create the structured markdown: {poster_id}_raw.md
4. Create the sub-JSON with poster content: {poster_id}_sub-json.json
5. Create the full JSON with complete metadata: {poster_id}.json

---

License

These annotations are provided for research and development purposes. Individual poster content remains subject to the original authors' rights and conference publication policies.