Dataset Source?
It is an amazing dataset! How was it made? I see no paper corresponding to it. This dataset does not seem to be related to UniChart pre-training or fine-tuning.
Thanks for your attention.
Hello, thanks for your interest in our work.
This dataset is just a derivative (reformatting) of the original UniChart dataset.
Q1: So is it a derivative of ChartQA (an unrevealed version)? I am surprised that the images are from the ChartQA dataset, however, the QA pairs are new.
Q2: Are the answers labeled by human annotators or generated by QA models?
Many thanks for your time!!!
No, it's a derivative of the UniChart dataset which is already publicly available:
https://huggingface.co/datasets/ahmed-masry/unichart-pretrain-data
https://huggingface.co/datasets/ahmed-masry/UniChart-pretrain-images
The UniChart has many images from the ChartQA dataset, but with different annotations (e.g., new QA pairs, new captions) for training the models.
Got it. Thank you~