Mengye Ren comments

Results 27 comments of


                                            Mengye Ren

Execution Error: The method 'ExecuteMso' from the object '_CommandBars' has failed

Same issue here.

Network architecture

We use the same architecture as used in Snell et al. Prototypical Networks 2017

Which split files should I use to generate data exactly the same as the npy dataset to tiered?

Hi `train.csv` is for train `val.csv` is for val `test.csv` is for test.

Labels of tieredImageNet

The labels are stored in the pickle (.pkl) file, whereas the images are stored in the .npz file.

Labels of tieredImageNet

Ok I have identified the problem. Sorry about that. Will update the dataset file tomorrow and update this thread. Thanks for reporting.

Labels of tieredImageNet

Hi, I updated the tar file with labels file. Please verify. Thanks.

What does m_dist_1 += tf.to_float(tf.equal(m_dist_1, 0.0)) mean?

This is to prevent it to be zero. So it will be changed to 1.0 when it's 0.0

New text dataset

Yes, I think you can use this to train your classifier to learn a semi-supervised algorithm, assuming that your dataset is also few-shot based. Sent from my iPhone > On...

New text dataset

Sorry, so it means you probably have many classes to sample few-shot episodes from.

Why using the labeled and unlabeled split?

We have a split in labeled and unlabeled because it is similar to the regular notion of semi-supervised learning. Using the rest of the query set is definitely possible, which...