WS_DAN
WS_DAN copied to clipboard
Why are pooled features multiplied by 100?
In bilinear_attention_pooling
the pooled features are multiplied by 100. Why is that?