API Reference#

This section provides comprehensive class, method, and function documentation for all modules in REANIMATOR-VLM.

Data Models (`reanimator.models`)#

Class	Description
`Document`	Top-level container representing a processed document.
`Table`	Extracted table with DataFrame content, caption, and position.
`Figure`	Extracted figure with caption, image path, and position.
`Formula`	Extracted math formula or equation.
`Topic`	Retrieval query topic object.
`Judgement`	Synthetic 4-level relevance judgment object.

Document Pipeline (`reanimator.pipeline`)#

Class	Description
`ReanimatorVLM`	Main document processing pipeline entry point.
`PDFLocalizer`	Bounding box normalization & figure image cropper.

Project Collection Workspace (`reanimator.collection`)#

Class / Method	Description
`ProjectCollection`	Workspace directory manager.
`ProjectCollection.process`	Stream and save parsed documents directly to disk.
`ProjectCollection.iter_documents`	Yield document instances lazily from disk.
`ProjectCollection.run_relevance_assessment`	Evaluate synthetic relevance and export TREC qrels.

Synthetic Relevance Assessment (`reanimator.relevance`)#

Class / Function	Description
`RelevanceEvaluator`	Core UMBRELA-style relevance judging engine.
`ModalityExtractor`	Extract target modalities (`full_document`, `text_only`, `tables`, `figures`, `formulas`, `chunks`).
`TopicAdapter`	Normalize query strings, lists, or TREC topic dicts.
`export_qrels`	Export judgments in standard TREC qrels format.

VLM Backends (`reanimator.backends`)#

Class	Description
`OpenAIVisionBackend`	OpenAI API & remote `dots.mocr` vLLM backend.
`GeminiBackend`	Google Gemini 2.0 Flash / Pro vision backend.
`CachedBackend`	Disk-backed response caching wrapper.

Visualization (`reanimator.visualization`)#

Function	Description
`draw_bounding_boxes`	Render color-coded bounding boxes and centered label pills onto page images.