This repository contains an unofficial implementation of "Conditional Object-Centric Learning from Video" (ICLR 2022). Note: In this implementation, only unconditional slot intialization is considered.

Paper Link: https://arxiv.org/abs/2111.12594

Clone this repository

git clone https://github.com/a-imamshah/savi-pytorch cd savi-pytorch

Create new environment

conda create --name savi python=3.8 source activate savi

Install all the dependencies

pip install -r requirements.txt

python train.py

Modify SAViParams in params.py to modify the hyperparameters.

To log outputs to wandb, run wandb login YOUR_API_KEY and set is_logging_enabled=True in SAViParams.

Credits to the original authors of the paper: Thomas Kipf, Gamaleldin F. Elsayed, Aravindh Mahendran, Austin Stone,Sara Sabour, Georg Heigold, Rico Jonschkowski, Alexey Dosovitskiy & Klaus Greff.

I adapted this code from the unofficial implementation of "Object-Centric Learning with Slot Attention" by Untitled AI.