search

Pang-Yatian / Point-MAE

[ECCV2022] Masked Autoencoders for Point Cloud Self-supervised Learning
MIT License
448 stars 55 forks source link

A question about "Few-Shot". #14

Closed QWTforGithub closed 2 years ago

QWTforGithub commented 2 years ago

Thank you very much for your help last time. I have a question about "fee-shot ". In the training set," 5way-10shot " means that we randomly selects 5 classes and has 10 samples for each class. May I ask how to select classes and samples in the test set? Is it still the same 5 classes with10 samples ? Or all of the samples for each class?

Pang-Yatian commented 2 years ago

Hi,

You may refer to our paper, in which the setting of few-shot learning is described in detail.

The short answer is that we select 20 UNSEEN samples for each of the 5 classes.

QWTforGithub commented 2 years ago

Hi,

You may refer to our paper, in which the setting of few-shot learning is described in detail.

The short answer is that we select 20 UNSEEN samples for each of the 5 classes.

Sorry, I found the answer in your article shortly after posting the question.

Pang-Yatian commented 2 years ago

Never mind.

kabouzeid commented 2 years ago

I have another question: in the paper you report mean accuracies with std from 10 independent experiments. Are those w.r.t. the final epoch or best epoch of each of the 10 experiments?

Pang-Yatian commented 2 years ago

I have another question: in the paper you report mean accuracies with std from 10 independent experiments. Are those w.r.t. the final epoch or best epoch of each of the 10 experiments?

Hi,

We report the best epoch results.