Dynamic-Few-Shot-Visual-Learning-without-Forgetting

pytorch simple implement for the CVPR 2018 paper Dynamic Few Shot Visual Learning without Forgetting in jupyter version.

Set up the dataset

download the MiniImagenet dataset from here and modify the data root.

Train the model

• Step 1 refers to pretrain Feature Extractor and cosine-based Classifier. In this process I set the epoch_num to 30.
• Step 2 refers to continue train Classifier and attention-based Few-shot Weight Generator.

Evaluate

• I get similar result as published in paper.