ronghanghu / tensorflow_compact_bilinear_pooling

Compact Bilinear Pooling in TensorFlow
Other
141 stars 45 forks source link

sum_pool #9

Open ahuxjz opened 4 years ago

ahuxjz commented 4 years ago

Hello, what is the function of the parameter sum_pooling in the code? If true seems to be a summation operation, is there any difference in principle if no summation is made? I don't quite understand, thank you.

ronghanghu commented 4 years ago

There isn't a difference in principle. The summation operation is used for output in Compact Bilinear Pooling (http://arxiv.org/abs/1511.06062, see their Figure 1), so I also added this option.

ahuxjz commented 4 years ago

原则上没有区别。求和运算用于紧凑型双线性池(http://arxiv.org/abs/1511.06062,请参见图1)中的输出,因此我也添加了此选项。 I see that the bilinear pooling is mostly used for the final fully connected layer. Then, if I only combine two feature graphs, the output dimensions are (batch_size, h, w, dim) instead of (batch_size, dim). I don't use sum_pooled, does this have any effect on the effect?

SWShao commented 4 years ago

Hello! I also want to combine two feature graphs without adopting sum_pooled. Do you try it or manage to do it?