ICDAR 2026, Vienna (Austria), Aug 31 - Sep 2
The evaluation process according to the 4 shared-tasks is given below:
Given a set of figure images, identify the class/type of the figure that is one of the 47 classes (visualization). example: bar chart, scatter plot, spectra chart, etc.
The performance was assessed using the following metrics:
Accuracy
Precision
Recall
F1-score
Given a set of quantitative-plot figures, extract the underlying data table in the standard markdown format. example:
| First Header | Second Header |
| --- | --- |
| Content Cell | Content Cell |
| Content Cell | Content Cell |
The Performance will be assessed using the following metrics:
RMS (Relative Mapping Similarity) (Liu et al. 2023)
TEDS (Tree Edit Distance Similarity) (Smock et al, 2021)
Given a set of quantitative-plot figures (and optionally caption and text content), extract an insightful summary. example:
"The chart illustrates a timeline of a single ALE cycle showing the values of the following process parameters: DC power, ICP power, CF4+H2 flow rate, and Ar flow rate."
The Performance will be assessed using the following metrics:
ROUGE (Lin et al, 2004)
BERTScore (Zhang et al, 2019)
Given a set of quantitative-plot figures (and optionally caption and text content), perform visual question-answering. example:
Q: What are the three main steps of this ALE cycle and what function fulfills the last sub-cycle?
A: The three main parts in this ALE cycle are the dosing of a plasma containing Ar, CH4, and H2 related species, a plasma exposure, solely consisting of Ar, and lastly a Ar plasma exposure with a biased substrate. This last step is introduced to enable energetic ions to bombard the substrate, leading to selective etching.
The Performance will be assessed according to each answer-type using the following metrics:
Paragraph - ROUGE and BERTScore
Factoid - Exact Match (EM) and ROUGE
Yes/No - Accuracy, Precision, Recall, F1-score
List - Set-based Precision, Recall, F1 (unordered lists)