Introduction to VIS30K:

VIS30K is a collection of 31,481 images including every figure and table for 30 years spanning each track of the IEEE Visualization conference series (Vis, SciVis, InfoVis, and VAST). VisImageNavigator now extends it through 2025 with the 2020 short papers and every journal and short paper of VIS 2021-2025, for a total of 39,339 figures and tables (plus 15,105 equations and 1,449 algorithms) from 1990 to 2025.

How are the images distributed over the years?

What image (figure and table) data can you find here?

Images from every year of the IEEE VIS conferences:
- VAST: 2006-2020
- InfoVis: 1995-2020
- Vis: 1990-2013
- SciVis: 2012-2020
- VIS short papers: 2020
- VIS: 2021-2025 (Starting in 2021, the three tracks (VAST, InfoVis, and SciVis) were merged into a single track.)

What Data are Released for Reproducible Research?

- 31,481 images in VIS30K stored in IEEE dataport.
- The meta data stored in google spread sheet.
- 10K Training and 3K validation datasets used in our CNN algorithms
- The image corpus and the text corpus for pseudo paper generation.
* For source code, please contact Prof. Jian Chen at chen.8028@osu.edu.

How to extract images automatically?

We used a CNN-based solution and end-to-end framework to extract figures and tables in research paper pages. The main idea behind our approach is to train a CNN with synthesized dummy papers, created with existing visualization image corpora. Equipped with the resulting training set creation, we then automatically extracted figures from paper pages. We have used Faster-RCNN and YOLOv3 to reduce the subsequent human effort. Please try our pre-trained model here.

How to cite this work?

For dataset, please cite:
Jian Chen, Meng Ling, Rui Li, Petra Isenberg, Tobias Isenberg, Michael Sedlmair, Torsten Möller, Robert S. Laramee, Han-Wei Shen, Katharina Wünsche, and Qiru Wang. VIS30K: A Collection of Figures and Tables from IEEE Visualization Conference Publications. IEEE Transactions on Visualization and Computer Graphics (2021).