yujiang01 / n2nmn_pytorch Goto Github PK

Thanks for this brilliant work. However, as I trained with the three settings, The performances (55 trained from scratch, 73 trained with gt_layout, 75 trained with gt+rl) could not reach those claimed in the original paper (69 trained from scratch, 79 trained with gt_layout, 84 trained with gt+rl). Did you reach the performances claimed in that paper? Thanks!