CACTUs-ProtoNets: Clustering to Automatically Generate Tasks for Unsupervised Prototypical Networks.
This code was used to produce the CACTUs-Protonets results and baselines in the paper Unsupervised Learning via Meta-Learning.
This repository was built off of Prototypical Networks for Few-Shot Learning.
The code was tested with the following setup:
- Ubuntu 16.04
- Python 3.6.6
- PyTorch 0.4.0
- Install PyTorch and torchvision.
- Install torchnet by running
pip install git+https://github.com/pytorch/tnt.git@master.
- Install the protonets package by running
python setup.py installor
python setup.py develop.
- Install scikit-learn.
The Omniglot splits with ACAI and BiGAN encodings used for the results in the paper are available here. Download and extract the archive's contents into this directory.
Unfortunately, due to licensing issues, I am not at liberty to re-distribute the miniImageNet or CelebA datasets. The code for these datasets is still presented for posterity.
You can find examples of scripts in
/scripts. All results were obtained using a single GPU.
The unsupervised representations were computed using three open-source codebases from prior works.
- Adversarial Feature Learning
- Deep Clustering for Unsupervised Learning of Visual Features
- Understanding and Improving Interpolation in Autoencoders via an Adversarial Regularizer
To ask questions or report issues, please open an issue on the issues tracker.