REANIMATOR-VLM Technical Documentation & API Reference

Welcome to the official technical documentation for REANIMATOR-VLM — a Python toolkit for multi-modal document parsing, visual layout grounding, disk-backed workspace management, and synthetic relevance assessment.

API Reference

• API Reference
  • Data Models (reanimator.models)
  • Document Pipeline (reanimator.pipeline)
  • Project Collection Workspace (reanimator.collection)
  • Synthetic Relevance Assessment (reanimator.relevance)
  • VLM Backends (reanimator.backends)
  • Visualization (reanimator.visualization)
• Core Data Models (reanimator.models)
  • Document
  • Table
  • Figure
  • Formula
  • Topic
  • Judgement
• Document Pipeline (reanimator.pipeline)
  • ReanimatorVLM
  • PDFLocalizer
• Project Collection Workspace (reanimator.collection)
  • ProjectCollection
• Synthetic Relevance Assessment (reanimator.relevance)
  • RelevanceEvaluator
  • ModalityExtractor
  • export_qrels
• VLM Backends (reanimator.backends)
  • OpenAIVisionBackend
  • GeminiBackend
  • CachedBackend
• Visual Bounding Box Rendering (reanimator.visualization)
  • draw_bounding_boxes

1. Overview & Architecture

REANIMATOR-VLM revitalizes scientific and complex document collections for Information Retrieval (IR) evaluation and Retrieval-Augmented Generation (RAG).

2. Quick Installation

pip install reanimator-vlm
pip install "reanimator-vlm[all]"

3. Quickstart Example

from reanimator import (
    ReanimatorVLM,
    ProjectCollection,
    OpenAIVisionBackend,
    CachedBackend,
    RelevanceEvaluator,
)

backend = CachedBackend(
    OpenAIVisionBackend(
        model="rednote-hilab/dots.mocr",
        base_url="http://139.6.160.244:6543/v1",
        api_key="not-needed",
        use_structured_layout=True,
    ),
    cache_dir="./.md_cache",
)
pipeline = ReanimatorVLM(backend=backend)

project = ProjectCollection(
    project_dir="./project_data/my_collection",
    arxiv_ids=["2504.07584"],
    topics=[{"topic_id": "101", "title": "Table parsing performance"}]
)

project.process(pipeline)

evaluator = RelevanceEvaluator(backend=backend, model_name="dots.mocr")
project.run_relevance_assessment(evaluator, modality="tables")