dvlab-research/PFENet

PFENet++

ily666666 opened this issue · 2 comments

PFENet++

However, these extreme values are vulnerable to the noisy support features that are locally identical to the query features but are actually from distinct classes.
在PFENet ++中的一句话。
support feature经过 mask操作之后剩下的部分不是只有相同类别的特征了吗,为什么还会有 distinct classes?

Thanks for being interested in our work.

Here we refer to the noisy foreground support features that have high correlation values with distinct classes in the query features such as the objects in the background of the query sample.