Abyss

Training Le continued to decline. Before the 6500th iteration, when the validate function is called, the LE of the test is falling, while the recall is rising, and everything is...

Hi,Yang!I see your model_attention code,in function xattn_score_t2i,you use similarities.append(row_sim+0.5*max_attn) to compute.I want to know why you use 0.5*max_attn,and whether you mention this in your paper.