Question about augmentation training data in FVQA 2.0

[zhuweihao12138]

Thank you very much for your work on FVQA 2.0.

I noticed that the augmentation training data is not included in the released resources here: https://github.com/LinWeizheDragon/Retrieval-Augmented-Visual-Question-Answering/blob/main/FVQA2.0.md. May I ask if you could publicly release the augmentation training data?

From the paper, my understanding is that:

• The 2,820 adversarial samples (around 1,376 questions per split on average) are exclusively used for testing.
• The augmentation data (around 2,262 questions per split on average) is used for training.

Could you please confirm if this is correct?

Thank you very much for your time and consideration.