Problem on the visualization about r2gen HOT 5 CLOSED

Dear @linzhlalala,

Thanks for your attention to our paper!

In our experiment, we visualize the attention weights of the first cross attention layer.

To deal with the multiple heads, we select the attention weights from the head which has the largest differences in the attention values.

The visualization code will be cleaned and updated here in the future.

Best,
Zhihong

from r2gen.

Thanks for sharing.

from r2gen.

Dear @linzhlalala,

Thanks for your attention to our paper!

In our experiment, we visualize the attention weights of the first cross attention layer.

To deal with the multiple heads, we select the attention weights from the head which has the largest differences in the attention values.

The visualization code will be cleaned and updated here in the future.

Best,
Zhihong

Hi Zhihong,
Thank you for sharing your code. I am also interested in the Visualizations of image-text attention mapping part in the paper.
Have you cleaned the visualization code? I don't find a good way to visualize the transformer-based model.

from r2gen.

Hi, any progress?

from r2gen.

I don't need the code. The idea is helpful enough. Thanks.

from r2gen.