search

edgarschnfld / CADA-VAE-PyTorch

Official implementation of the paper "Generalized Zero- and Few-Shot Learning via Aligned Variational Autoencoders" (CVPR 2019)
MIT License
283 stars 57 forks source link

general question regarding training and testing #21

Closed seyeeet closed 4 years ago

seyeeet commented 4 years ago

Thanks for sharing your code. I am new to FSL and appreciate it if you can clarify this confusion for me. In FSL we have support and query images. Should we have all the supports and queries for each episode to be in one iteration?

edgarschnfld commented 4 years ago

Hi, there are different ways of conducting FSL. We are using the conceptually simplest setup: If you have K classes with N images each, and you train with S shots, then the training set for the classifier consists of KS images, while the test set consists of K(N-S) images. The classifier is only trained and test once.