VAE Embedding Visualization

A small, light and fast reactive web tool to quickly visualize and analyze VAE embeddings. A very useful tool for manually analyzing and evaluating trained VAE models. Being web-based, it allows quickly presenting and sharing results on the web without having the end user to deal with any code.

Quick Start

Client

git clone https://github.com/saravanabalagi/vae_visualization
cd vae_visualization
yarn install
yarn run build

Server

• Copy given abstract server.py file to your python codebase
  • Only very minimal changes to server.py file are needed if your project is based on PyTorch Lightning
  • The code can easily be adapted to load other Pytorch models and with a little more effort, you should be able to load tensorflow models as well just following the structure!
• Modify it to load your pretrained model from ckpt_path
• Keep all image directories under a single img_dir directory
• Make sure you build the client and modify VAE_VISUALIZATION_CLIENT_PATH to point to repo/public
• Run python server.py <ckpt_path> <img_dir> to start the server to load from ckpt

The visualization tool should now be available at http://localhost:5000 in your browser:

Note that this is intended for research and development purposes, this requires more tweaks if you plan to use it in production. This tool may not efficiently handle multiple image directories with more than tens of thousands of images due to the way it serves autoindex jsons.