Image & Text attention
SuryaThiru opened this issue · comments
I'm looking to visualize the attention heads of models like layoutlmv3 that taken in both the text tokens and image inputs in a meaningful way. Is there a way I can do this with bertviz?
Thank you!
Hi @SuryaThiru there may be a way to extend it somehow with the patch embeddings, but it isn't something I've looked at in detail. Sorry not to be more helpful here.