One-shot learning
Zero-shot learning
Multi-shot learning
Sparse
Fine-grained Fine-tune
Background: CVPR 2018 collection of four papers on the study of a small sample, but to CVPR 2019, that number soared to nearly 20
So what is the small sample learning it?
Learn inside the machine, when you have a lot of training samples for training, and if your test set and the training set is not the same, this time called the support set of support data. In the test, you will face a new category (usually 5 class), where each category only a very small sample (usually only one per class or five samples, called "support set"), as well as from the same query image category.
Next, this small sample method will be divided into five different categories (although these categories are not clearly defined, many methods belong to more than one category).
(1) measure learning
(2) meta-learning
(3) the data enhancement method