Zerox is a multimodal document parser and OCR tool that uses vision models to convert PDF files and images into structured Markdown text. It functions as a visual layout extraction engine, leveraging large multimodal models to digitize documents while maintaining their original structural formatting.
The system differentiates itself through the use of coordinate-based element mapping and multimodal layout analysis to identify structural elements like tables, charts, and headers. It utilizes rasterization to convert vector PDF pages into high-resolution bitmaps, ensuring consistent input for the vision models used to synthesize the final Markdown output.
The tool covers a broad range of document digitization capabilities, including complex layout extraction and vision-based OCR. It processes visual document representations to interpret the spatial relationship between text and data, converting them into machine-readable formats.
