Ai Dreams Forum

Member's Experiments & Projects => General Project Discussion => Topic started by: hainaa on April 22, 2019, 07:28:47 am

Title: VQA based on image captions
Post by: hainaa on April 22, 2019, 07:28:47 am

Hi,

Is there some paper related to the task of answering an image based question given as input an image and its description.
I tried to search online but I didn't find any but I remember some days back I saw such work. If anyone know of some work related to it, please share the link.

Thanks!

Title: Re: VQA based on image captions
Post by: LOCKSUIT on April 22, 2019, 08:46:11 am

like google deepmind or somethin? I seen something like white images with circles and patterns of problem solving in a video i seen online, on I think it was Two Minute Papers's youtube channel. They make it generally aswer questions like what is the most likely pattern that comes next in this array of circles...

Title: Re: VQA based on image captions
Post by: ivan.moony on April 22, 2019, 10:52:11 am

Is it this one?

Ai robotic news link (https://aidreams.co.uk/forum/index.php?topic=13896.msg57597#msg57597)

Homepage of CLEVR project (https://cs.stanford.edu/people/jcjohns/clevr/)

Title: Re: VQA based on image captions
Post by: hainaa on April 23, 2019, 05:59:01 am

No, It's different. It's great work but I want one where input is (question, image , image description) and output is the answer for the question.

Title: Re: VQA based on image captions
Post by: WriterOfMinds on April 23, 2019, 06:53:11 am

The closest thing I can think of is Visual Chatbot. It generates the image description rather than taking it as input, but I think it can use dialogue history (including the generated caption?) to help answer followup questions.

https://arxiv.org/abs/1611.08669

Title: Re: VQA based on image captions
Post by: hainaa on April 23, 2019, 06:57:59 am

Actually this is really interesting and maybe I could get ideas from it. Thanks for the link  :)