A question about mini-imagenet about maml HOT 10 CLOSED

Dear Chelsea,

following up on this discussion - In your paper you say that the split is 64 train / 12 validation / 24 test classes. The original split (https://github.com/twitter/meta-learning-lstm/tree/master/data/miniImagenet) is 64/16/20 though. Is that just a typo, or did you use a different split?

If anybody is still looking for the dataset, I found this source: dragen1860/LearningToCompare-Pytorch#4

from maml.

@jongchyisu @imaxpayne Hi, I have sent email to Sachin, but got no response, could you share the solution to this problem? Thanks a lot.

from maml.

Yes, it is a typo.

from maml.

Sorry this is unclear. I got the dataset from Ravi & Larochelle '17. If you email Sachin, he should be able to help.

from maml.

I updated the instructions in proc_images.py

from maml.

Thank you! I will contact him via email.

from maml.

Hi Chelsea,

Sorry for opening this issue again, but from the csv files in Sachin's GitHub, I can't find some of the images in the imagenet by the ID. For example, the first training image is n0153282900000005.jpg,n01532829
but there is no n01532829_5.jpg can be found in
http://www.image-net.org/api/text/imagenet.synset.geturls.getmapping?wnid=n01532829
How did you solve this issue? Thank you very much!

from maml.

@jongchyisu You should contact Sachin via email.

from maml.

Yes I just sent him an email.

from maml.

I also sent him an emial and there was no response. It would be great if you could provide that dataset.
@cbfinn

from maml.