Size of extracted features
jasonwei20 opened this issue · comments
Could you point me to where the code resizes the extracted features?
For ResNet-50, I got 9216 extracted features after layers 1, 2, and 4, and 8192 features after layers 3 and 5. Where do these numbers come from?
Thank you and apologizes if this is a silly question!
Nevermind, I see how it's done now.
To calculate the number of extracted features, note the spatial resolution X
, kernel size of avg pool layer K_d
, and stride of avg pool layer S_d
. Following Table 16 in the paper, I found that the downsampled spatial resolution X_d
is equal to (X-K_d+2*P_d) / S_d
, where P_d
is the padding which is mostly zero.
Closing this issue, thanks!