GuodongQi / CMVAE

Codes for "CMVAE: Causal Meta VAE for Unsupervised Meta-Learning"

CMVAE: Causal Meta VAE For Unsupervised Meta-Learning

This is the Pytorch implementation for the CMVAE

Dependencies

This code is written in Python. Dependencies include

• python >= 3.6
• pytorch = 1.4 or 1.7
• tqdm, wandb

Data

• Download Omniglot data from here.
• Download pretrained features for Mini-ImageNet from here.
• Download pretrained features for CelebA from here.
• (Optional) If you want to train SimCLR from scratch, download images for ImageNet from here amd CelebA here

data directory should be look like this:

data/

├── omiglot/
  ├── train.npy
  ├── val.npy
  └── test.npy
  
├── mimgnet/
  ├── train_features.npy
  ├── val_features.npy
  └── test_features.npy
 
├── celeba/
  ├── train_features.npy
  ├── val_features.npy
  └── test_features.npy
    
└── imgnet or celeba_imgs/ -> (optional) if you want to train SimCLR from scratch
  ├── images/
    ├── n0210891500001298.jpg  
    ├── n0287152500001298.jpg 
	       ...
    └── n0236282200001298.jpg 
  ├── train.csv
  ├── val.csv
  └── test.csv

Experiment

To reproduce Omniglot 5-way experiment for CMVAE, run the following code:

cd omniglot
python main.py --data-dir DATA DIRECTORY   --save-dir SAVE DIRECTORY  --way 5 --sample-size 200

To reproduce Omniglot 20-way experiment for CMVAE, run the following code:

cd omniglot
python main.py --data-dir DATA DIRECTORY   --save-dir SAVE DIRECTORY   --way 20 --sample-size 300

To reproduce Mini-ImageNet 5-way experiment for CMVAE, run the following code:

cd mimgnet
python main.py --data-dir DATA DIRECTORY   --save-dir SAVE DIRECTORY  

To reproduce CelebA 5-way experiment for CMVAE, run the following code:

cd celeba
python main.py --data-dir DATA DIRECTORY   --save-dir SAVE DIRECTORY  

(Optional) To reproduce SimCLR features for Mini-ImageNet, run the following code:

cd simclr
python main.py --data-dir DATA DIRECTORY  --save-dir SAVE DIRECTORY   --feature-save-dir FEATURE SAVE DIRECTORY  

Acknowledgments

Our work and code benefit from two existing works, which we are very grateful.
Meta-GMVAE
notears