Competition on HArvesting Raw Tables (CHART) 2019 - PubMedCentral (ICDAR-CHART-2019-PMC)

Research Tasks

Legend Analysis

2019-06-18 (v. 1)

Contact author

Kenny Davila

University at Buffalo


This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.


The purpose of chart legends is to associate a data series name with the graphical style used to represent it This is critical to chart understanding when there are multiple data series represented.

Competing systems are expected to associate each legend label text with the corresponding graphical style element within the legend area. Bounding boxes and transcriptions (but not text roles) are given as input. Note that in this task, legend labels are not paired with the corresponding data series found in the plot area. Also, some charts do not have legends, and an empty list should be returned.


For each GT legend label, if there is an associated predicted graphical style element, we compute the IOU of the predicted BB to the GT graphical style element BB. We then divide the sum of the IOU by max(#predicted, #GT) for each image, and then average this value over all images.

For charts that have no legend, it is expected that participant systems return an empty list to receive the max score for that chart. When there is no legened, specifying any output results in a score of 0 for that chart.


No comments on this dataset yet.

Add your comment

In order to comment on a dataset you need to be logged on
Register Now!


In order to rate this dataset you need to be logged on
Register Now!