Closed jychoi118 closed 4 years ago
Hi, @jychoi118,
I think in general it's true that better models lead to higher accuracy, and this is true for the progress of few-shot recognition in past years.
For the ResNet-50 on MiniImageNet, I feel it might be an overkill for such as small data set, and it might be related to "Deep Double Descent" phenomenon (see here).
One potential solution to fix this might be using more stronger data augmentation for ResNet-50, which should be fun to see but is not the point of the project.
In table 3 of your paper, "supervised" model uses ResNet-50 as backbone, with accuracy of 73.81% in miniImagenet 5-way. However, "Ours-simple" model in table 1 with ResNet-12 reports 79.64%.
But as you mentioned in section 4.9, results should improve with better backbone networks.
Can you explain why the accuracy with ResNet-50 is lower than ResNet-12?