disent 0.0.1.dev1

Vae disentanglement framework built with pytorch lightning.

🧶 Disent ⚠️ [W.I.P]

Disentanglement Library for pytorch and pytorch-lightning. With an easy to use configuration based on Hydra.

Documentation: Check out the docs and examples!

Another disentanglement library?

• I needed to become more familiar with VAE's (Currently working on my masters)

• DISCLAIMER: This project has its roots in the tensorflow disentanglement_lib library.

• Weakly-Supervised Disentanglement Without Compromises stated they would release their code as part of disentanglement_lib... I didn't have time to wait... As of September it has been released.

• The disentanglement_lib still uses Tensorflow 1.0 and Gin Config controls execution, hiding the flow of data in the library (I am not a fan).

Features

Frameworks

• Unsupervised:
  • VAE:
  • BetaVAE:
  • DFCVAE:
• Weakly Supervised:
  • Ada-GVAE: AdaVae(..., average_mode='gvae')
  • Ada-ML-VAE: AdaVae(..., average_mode='ml-vae')
• Supervised:
  • TVAE:

Metrics

• Disentanglement:
  • FactorVAE score:
  • DCI:

Datasets:

• Ground Truth:
  • Cars3D:
  • dSprites:
  • MPI3D:
  • SmallNORB:
  • Shapes3D:
• Ground Truth Non-Overlapping (Synthetic):
  • XYBlocks: 3 blocks of decreasing size that move across a grid. Blocks can be one of three colors R, G, B. if a smaller block overlaps a larger one and is the same color, the block is xor'd to black.
  • XYSquares: 3 squares (R, G, B) that move across a non-overlapping grid. Obervations have no channel-wise loss overlap.
  • XYObject: A simplistic version of dSprites with a single square.

Usage

Disent is still under active development (I an sorry there are no tests yet).

The easiest way to use this library is by running experiements/hydra_system.py and changing the config in experiements/config/config.yaml. Configurations are managed by Hydra Config