[Asking about out-of-context concept]

[Trọng Đạt Phan]

Dear Deep Fake challenge organizers,

Our team is joining this challenge and after considering some samples from the public test set, we have some questions about the definition of out-of-context, which is needed your clarification.

1. If two captions are not relating, can we consider this case as out-of-context, without paying attention to the relationships between two captions and the image. 
2. If two captions have a similar meaning or relating to each other, but both of them are not relevant to the given image. Can this case be treated as out-of-context?

Thank you very much for answering our questions. 

Best regards,
--

Trong-Dat Phan

Mobile: +84 34 618 2803
Email: phant...@gmail.com
Skype: live:phantrdat

[shivangi.tum]

Hi,

From the paper "We define out-of-context use of images as presenting the image as evidence of untrue or unrelated event(s). If the two captions refer to the same object in the image but are semantically different, i.e. correspond to different events, then it indicates out-of-context use of the image. However, if the captions correspond to the same event irrespective of the object(s) the captions describe, it is defined as not-out-of-context."

Refer to the paper for more details: https://arxiv.org/abs/2101.06278

To clarify, for this task, we consider the relationship of both the captions with the provided image and not between the captions.

The GT labels are already provided for the test dataset. Please use them for evaluation. You can design the algorithm any way you like, but please make sure to have the same setup for all the samples.