0

I am trying to implement the paper Beyond triplet loss: a deep quadruplet network for person re-identification. In the paper, they provide a figure (attached below) containing the network architecture, but they don't specify the exact kernel sizes.

enter image description here

Can anyone help with getting the exact network architecture? Thanks in advance.

Robin van Hoorn
  • 1,810
  • 7
  • 32
Varghese Kuruvilla
  • 1

1 Answers1

1

Welcome to AI SE! I think they use an AlexNet architecture for the following 2 reasons.

1: The AlexNet architecture matches exactly with the architecture shown in Figure 3 (5 conv layers and 3 fc layers).

2: Looking at the paper, they mention

For all other networks, we use a pretrained AlexNet model

Although they do not specifically state that they use AlexNet for their own work, this was (and is) very common for representation learning papers. In addition, if you use AlexNet for the baseline models, it is very logical to take the same architecture for your new model.

Kernel sizes/strides: Seeing both their reference to AlexNet and the matching architecture, I would assume they use AlexNet for their own architecture as well. Please see the AlexNet wikipedia figure for an easy overview of the kernel-sizes/strides of AlexNet.

Robin van Hoorn
  • 1,810
  • 7
  • 32