search

yinboc / few-shot-meta-baseline

Meta-Baseline: Exploring Simple Meta-Learning for Few-Shot Learning, in ICCV 2021
MIT License
613 stars 107 forks source link

Prototypical Networks? #4

Closed hsqzzpf closed 4 years ago

hsqzzpf commented 4 years ago

Hi,

Thanks for your work.

Recently I am reading the paper "Prototypical Networks". I found your idea is quite similar to that one. Could you help explain a little bit about the main difference between your idea and that one?

What will the accuracy of ProtoNets on miniImageNet be if the backbone function is changed to ResNet that used in your paper?

yinboc commented 4 years ago

You can check Appendix B for details.

For the results, the original implementation of ProtoNet on miniImageNet is about 53.77 / 75.71 for 1-shot / 5-shot. Table 6 also demonstrates a comparison to a "improved ProtoNet" (with classification pre-training and introducing the temperature on logits).

hsqzzpf commented 4 years ago

Thanks for your prompt response. That helps a lot!

yinboc commented 4 years ago

You're welcome.