One shot learning

One single image
(neural networks don’t work well given one training example)
(Such a small training set is really not enough to train a robust neural network)
(If softmax, need to retrain every time new class adds in)

Solution:
Siamese Network

To train with triplet loss, every class in training set need to have at least two examples.
Choose triplets that are hard to train on.
Variants of Siamese Network

标签：training,set,shot,Network,Siamese,train,learning,need
来源： https://blog.csdn.net/weixin_42388228/article/details/90402264