Wsi-Caption What is text?

Hi, authors! Thanks for your outstanding work!

In the framework diagram of your paper, what does the Text in the lower right corner refer to?

Looking forward to your reply!

Best regards, Suoni

Nov 20 '24 13:11 lsnls

Just like transformer, you need to feed 'text' to the decoder when training.

Nov 21 '24 06:11 cpystan

hi, @cpystan , thank you for your quick response!

Your reply seems like explain the training paradigm of Text. I am still confused whether Text refers to a prompt or a question? And what is its specific content?

Thank you again.

Nov 22 '24 05:11 lsnls

For inference, special token of [START] is fed to the decoder so as to achieve autoregressive generation.

Nov 22 '24 07:11 cpystan

Thanks for your quick response!

You fed Text to the decoder during training. Was it only Text for all samples, and not any other specific sentences? Is that what you mean?

Nov 23 '24 12:11 lsnls