Reproduce the Visualization of Attention States

[peterfind]

Hello!

Thanks for this great code!

I find there is the visualization of attention states for captions, like this.

I want to reproduce the visualization results, but I cannot find the corresponding code in this repo.

I read the paper (Axiomatic attribution for deep networks) that you cited, and I think it's not easy for me to reproduce the visualization results correctly.

May I have the code to produce the visualization results, or is there anything I missed?

Thank you very much!

[Vincent-Ww]

I also want to do it but fails.
have you reproduced the visualization results successfully? Thanks

[rongtongxueya]

has the same question