poster2json extracts structured metadata from scientific conference posters (PDF or image format) into machine-actionable JSON conforming to the poster-json-schema.
The pipeline uses:
- Llama-3.1-8B-Instruct (a verbatim mirror of Meta's release; swap with any HuggingFace instruct model via
--model) for JSON structuring - pdfplumber for layout-aware PDF text extraction (stream-order
altolayout engine by default;POSTER2JSON_LAYOUT=blocksselects the XY-cut engine) - lingua-language-detector for ISO 639-1 language detection on body text (overrides any value the model emits — body text beats metadata-fragment guessing)
- ROR (
https://api.ror.org) for affiliation canonicalisation; matched names get a ROR identifier attached - ORCID (
https://pub.orcid.org) for author disambiguation; matched creators get an ORCID identifier attached (uses the authenticated Member API whenORCID_CLIENT_IDandORCID_CLIENT_SECRETare set, otherwise the public API)
pip install poster2json# Extract metadata from a poster (default: Llama-3.1-8B-Instruct @ 4bit)
poster2json extract poster.pdf -o result.json
# Use a different instruct model (any HuggingFace repo id works)
poster2json extract poster.pdf --model google/gemma-2-9b-it --quantization 4bit
# Trade VRAM for quality
poster2json extract poster.pdf --quantization 8bit
poster2json extract poster.pdf --quantization fp16
# Validate extracted JSON
poster2json validate result.json
# Process multiple posters
poster2json batch ./posters/ -o ./output/from poster2json import extract_poster, validate_poster
# Extract metadata
result = extract_poster("poster.pdf")
print(result["titles"][0]["title"])
# Validate the result
is_valid = validate_poster(result)Output conforms to the poster-json-schema (DataCite 4.7):
{
"$schema": "https://posters.science/schema/v0.2/poster_schema.json",
"creators": [
{
"name": "Garcia, Sofia",
"givenName": "Sofia",
"familyName": "Garcia",
"affiliation": [
{
"name": "Stanford University",
"affiliationIdentifier": "https://ror.org/00f54p054",
"affiliationIdentifierScheme": "ROR",
"schemeURI": "https://ror.org/"
}
]
}
],
"titles": [
{ "title": "Machine Learning Approaches to Diabetic Retinopathy Detection" }
],
"publicationYear": null,
"language": "en",
"researchField": "Health Sciences",
"subjects": [
{ "subject": "Machine Learning" },
{ "subject": "Diabetic Retinopathy" }
],
"descriptions": [
{ "description": "We present a deep learning model...", "descriptionType": "Other" }
],
"publisher": null,
"content": {
"sections": [
{ "sectionTitle": "Abstract", "sectionContent": "..." },
{ "sectionTitle": "Methods", "sectionContent": "..." },
{ "sectionTitle": "Results", "sectionContent": "..." }
]
},
"imageCaptions": [{ "id": "fig1", "caption": "Figure 1. ROC curves showing..." }],
"tableCaptions": [{ "id": "table1", "caption": "Table 1. Performance metrics" }],
"formats": ["application/pdf"]
}Notes on the auto-populated fields:
languageis detected from the raw body text (lingua heuristic). Returns null when text is too short (<200 chars / <50 non-ASCII codepoints) or the detector is unsure.researchFieldmust be one of the four OpenAlex top-level domains:Health Sciences,Life Sciences,Physical Sciences,Social Sciences. Null when the model can't pick one confidently.affiliationgets ROR enrichment when the matcher returns a high-confidence chosen result. Strings without a confident match pass through unchanged. SetPOSTER2JSON_ROR=0to disable.publisherandpublicationYearare always emitted asnull. They are platform-owned and set when the poster is published, not by extraction.formatsis derived from the input file's extension, not the model.
| Requirement | Specification |
|---|---|
| GPU | NVIDIA CUDA-capable, ≥8GB VRAM (default 4bit); ≥16GB for --quantization fp16 or image/OCR posters |
| RAM | ≥32GB recommended |
| Python | 3.10+ |
| OS | Linux, macOS, Windows (via WSL2) |
Field accuracy of poster2json 0.10.5 at default settings on 21 hand-labeled posters (20 PDF, 1 image). Each field is compared with the text printed on the poster using ROUGE-L, where 1.00 means identical. Posters is the number of labeled posters that have the field.
| Field | Posters | Mean ROUGE-L |
|---|---|---|
| Title | 21 | 0.978 |
| Author names | 21 | 0.898 |
| Given/family split | 12 | 1.000 |
| Affiliations | 21 | 0.953 |
| Author-affiliation assignment | 21 | 0.884 |
| Section text | 21 | 0.839 |
| Figure captions | 13 | 0.975 |
| Table captions | 5 | 1.000 |
| All content (forgiving) | 21 | 0.862 |
| Content, boundary-free F1 | 21 | 0.917 |
Only fields copied from the poster are scored. Summaries, keywords and the research field are written or chosen by the model, so they have no score. See Evaluation for how each field is compared.
| Document | Description |
|---|---|
| Architecture | Technical details & methodology |
| Evaluation | Validation metrics & results |
# Clone the repository
git clone https://github.com/fairdataihub/poster2json.git
cd poster2json
# Create a virtual environment
python -m venv .venv
# Activate the virtual environment
source venv/bin/activate
.venv\Scripts\activate # On Windows
# Install poetry
pip install poetry
# Install dependencies
poetry install
# Run tests
poe test
# Format code
poe formatIf you are on windows and have multiple python versions, you can use the following commands:
py -0p # list all python versions
py -3.12 -m venv .venvMIT License - see LICENSE for details.
@software{poster2json2026,
title = {poster2json: Scientific Poster to JSON Metadata Extraction},
author = {O'Neill, James and Soundarajan, Sanjay and Portillo, Dorian and Patel, Bhavesh},
year = {2026},
version = {0.8.0},
url = {https://github.com/fairdataihub/poster2json},
doi = {10.5281/zenodo.18320010}
}This project is funded by The Navigation Fund (10.71707/rk36-9x79).
Contributions welcome! Please see CONTRIBUTING.md for guidelines.
