The reproduction code for the "Continual Adaptation of Visual Representations via Domain Randomization and Meta-learning". Unofficial. This repo only contain the digit datasets part, with 4 digit datasets and the resnet-18 model, and its training testing code.
I implement the MetaLearning step by using the python package 'learn2learn'. see https://github.com/learnables/learn2learn
To run this repo, first you need to download 4 datasets. MNIST, MNIST-M, SYN, SVHN. Put them in the data folder. Then pip install learn2learn. Run main.py. That's all, the code is simple.