Skip to content

TF-ID

Detect tables and figures in academic papers

Visit official site

What TF-ID is

The model family is described as a set of object detection models by Yifei Hu that pull tables and figures out of academic papers. Versions with and without caption text are offered.

What you can do with it

  • Extract tables and figures from academic papers
  • Choose a version with or without caption text
  • Train a custom model with the provided guides

Application scenarios

  • A researcher extracts experiment result tables
  • A scholar analyses figure data in older literature
  • An education provider helps students find statistics

Pros & cons

✓ What we like

  • Custom model training is supported
  • Tables and figures are both handled

! What to watch out for

  • Use starts by cloning the GitHub repository
  • Annotation and image files must be placed in the required directories

FAQ

What is it based on?

The microsoft/Florence-2 checkpoint is named as the fine-tuning base.

What is released openly?

Model weights and a human-annotated dataset are listed.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review