Length of the Feature Vector for Pedestrian Representation

[WangLiwen1994]

To extract the feature vector of each input, the paper said you obtain a "1024-d vector as the final pedestrian representation". However, when I use the function "def extract_features(model, data_loader, print_freq=10):", I obtain a 12,288-d feature vector for each input image. So, I was wondering how to obtain the 1024-d representation in your code.

Thank you very much!