search

hysts / pytorch_mpiigaze

An unofficial PyTorch implementation of MPIIGaze and MPIIFaceGaze
MIT License
346 stars 85 forks source link

Questions about resnet-simple-14 #46

Closed Kelly-ZH closed 3 years ago

Kelly-ZH commented 3 years ago

Hi,hysts Firstly, thank you for your wonderful project, I can learn a lot from it. And when I was learning, I have several questions as follow, can you answer them for me?

  1. Why resnet-simple-14 has only one 11 convolution layer to learn spatial weight, and for the spatial weight CNN in the paper, it has three 11 convolution layers?
  2. I have trained resnet34 and resnet10 for the mpiifacegaze datasets, but they do worse than resnet14? What do you think of this result? Can you explain it to me?

Thank you for your reading, and I am really looking forward to your reply.

hysts commented 3 years ago

Hi, @Kelly-ZH

  1. That's because the performance didn't improve even with three convolutional layers.
  2. I'm not sure, but in general, a model with too many parameters could overfit.
Kelly-ZH commented 3 years ago

Hi, @Kelly-ZH

  1. That's because the performance didn't improve even with three convolutional layers.
  2. I'm not sure, but in general, a model with too many parameters could overfit.

Thank you very much.