About the grid feature extraction

[wangran1999]

Dear professor,
I 'm really confused about the grid feature extraction procedure. As far as I'm concerned, there should be two methods to extract grid features, one is 14×14Pool, and the other is 1×1 pool. But what makes me confused is that the two kinds of pooling seem to just be for region features, however, the grid feature are directly extracted from Resnet's c1 to c5. I don't know how the two kinds of pooling influence the extraction of grid feature. Could you please solve my confusion please? Thanks a lot.