Open ryzejiang opened 5 years ago
In paper, TUM(i)'s input is the channel of TUM(i-1)'s output[-1], and it is 128 in paper;
But in code, TUM's output[-1] channel is 256, and input(not first one) channel is 256+128(just like FFMv2 128 concat 256);
So,,, Why? Code and Paper is different.. Did I get it wrong?
In paper, TUM(i)'s input is the channel of TUM(i-1)'s output[-1], and it is 128 in paper;
But in code, TUM's output[-1] channel is 256, and input(not first one) channel is 256+128(just like FFMv2 128 concat 256);
So,,, Why? Code and Paper is different.. Did I get it wrong?