CSVaDE

Original implementation of Common Space Variational Deep Embedding in PyTorch.

Requirments

The code is tested with Python 3.8.3. A complete list of packages required is listed below:

Package	Version
pytorch	1.5.1
numpy	1.18.5
scikit-learn	0.23.1
tensorboard	2.0.0

Project Structure

├── README.md
│
├── configs
│   ├── embedding_size.json
│   ├── few-shot.json
│   ├── hidden_layers_1.json
│   ├── hidden_layers_2.json
│   ├── hidden_layers_3.json
│   ├── optimizer.json
│   ├── seen-unseen.json
│   └── zero-shot.json
│
├── datasets
│   ├── AWA1
│   ├── AWA2
│   ├── CUB
│   └── SUN
│
├── saved
│   └── csvade.pt
│
├── tensorboards
│   ├── experiments
│   └── models
│
├── main.py
├── models.py
├── train.py
├── data.py
└── utils.py

• configs\ contains architecture configurations for various experiements (hyperparameter tuning)
• datasets\ contains AWA1, AWA2, CUB and SUN datasets in pt (pytorch) file format
• saved\ contains pretrained models
• tensorboards\ contains tensorboard insights for pretained models
• *.py files contains the actual project code (start from main)

*Some of the above directories may not be present in the repo due to size restriction but will auto-generate when you run the code.

*Link to data folder will be provided soon.

Run

You can either run in single mode using a single set of options:

main.py --dataset AWA2 --num-shots 0 csvade

or in batch mode for hyperparameter turning

main.py --dataset AWA2 --num-shots 0 configs/embedding_size.json

for more options run:

main.py -h