Questions about Cross-Domain Few-Shot Learning (CD-FSL) Challenge
Closed this issue · 7 comments
Hallo, I see in your paper, the results of approach based Transfer learning is obviously better than the results of methods based Meta learning. I wonder if the challenge will evaluate the results of the two kind of methods separately.
And could we use a different backbone networks, such as ResNet18?
Hi @leezhp1994, you can propose your own methods for the challenge, it does not matter that whether it is transfer learning or meta-learning. And you can use any backbone.
Hi, in my opinion, the transfer learning methods shown in your paper actually use the samples from the target domain to fine-tune. However, the CD-FSL challenge requires us only to use the data from mini-imagenet, which means we cannot use any kind of fine-tune. Is that right?
@chrisyxue You can use fine-tune. The data from mini-imagenet is used for pre-training. You can use the support set of the target domain data for fine-tuning.
@chrisyxue You can use fine-tune. The data from mini-imagenet is used for pre-training. You can use the support set of the target domain data for fine-tuning.
So can we use the data from mini-imagenet when fine-tuning?
hi @yunhuiguo, can we use the data from mini-imagenet when fine-tuning?
hi @yunhuiguo, can we use the data from mini-imagenet when fine-tuning?
Yes.
hi @yunhuiguo, can we use the data from mini-imagenet when fine-tuning?
Yes.
When don't use any ground truth of query data, can we use the pseudo-label of query sets or access the query data when fine-tuning?