paper question

[Tracyxzh]

i have read your paper, i have some question:

1. the rmac or mac is one of code for a image?
2. why you discard all fully connect layer?
3. you did not train vgg16 or alexnet?

[gtolias]

1. Yes
2. It allows you to feed images of any dimensionality/aspect ratio. It is also translation invariant.
3. There is no training involved in this work/code. Only pre-trained networks are used. Fine-tuning for retrieval can be found here https://github.com/filipradenovic/cnnimageretrieval/