You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for your great work! I can easily visualize cross-attention map of each token in prompt on the origin image, but the self-attention visualization is a little bit confusing for me, since self-attention map represents the attention scores between image tokens, so it should be sized (H'W')(H'W'), so how could i map it back to HW and relfect it as shown in Fig4, thanks for your generous help!
The text was updated successfully, but these errors were encountered:
Thanks for your great work! I can easily visualize cross-attention map of each token in prompt on the origin image, but the self-attention visualization is a little bit confusing for me, since self-attention map represents the attention scores between image tokens, so it should be sized (H'W')_(H'W'), so how could i map it back to H_W and relfect it as shown in Fig4, thanks for your generous help!
Hi, have you resolve the problem? I want to know how to visualize the attn_map, too.
Thanks for your great work! I can easily visualize cross-attention map of each token in prompt on the origin image, but the self-attention visualization is a little bit confusing for me, since self-attention map represents the attention scores between image tokens, so it should be sized (H'W')(H'W'), so how could i map it back to HW and relfect it as shown in Fig4, thanks for your generous help!
The text was updated successfully, but these errors were encountered: