How do you understand the weight of the word 'woman' '0.54' or the word 'is' '0.37' in the paper: show, Attend and Tell?

[RunCode90]

Hi, have you read the paper: show, Attend and Tell:Neural image caption generation with visual attention. At last, the experimental results are visualized. For example: A(0.98) woman(0.54) is(0.37) throwing(0.33) a(0.28) frisbee(0.37) in(0.21) a(0.18) park(0.35) .(0.33).
How do you understand the weight of the word 'woman' '0.54' or the word 'is' '0.37',and have you achieved it in your code?

Any help will be highly appreciated. Thanks
Best wishs!

[giangnguyen2412]

Hi,

For a long time I have not touched its code so I can not remember exactly I have achieved what you said in or not.

In example: A(0.98) woman(0.54) is(0.37) throwing(0.33) a(0.28) frisbee(0.37) in(0.21) a(0.18) park(0.35) .(0.33). I think the scores is the attention weights that a region has at inference.

[RunCode90]

Thank you. I will try.
Best wishs!