Identifiability Results for Multimodal Contrastive Learning

Official code for the paper Identifiability Results for Multimodal Contrastive Learning presented at ICLR 2023. This repository contains pytorch code to reproduce the numerical simulation and the image/text experiment. The code to generate the image/text data from scratch is provided in a separate repo: Multimodal3DIdent.

Installation

This project was developed with Python 3.10 and PyTorch 1.13.1. The dependencies can be installed as follows:

# install dependencies (preferably, inside your conda/virtual environment)
$ pip install -r requirements.txt

# test if pytorch was installed with cuda support; should not raise an error
$ python -c "import torch; assert torch.cuda.device_count() > 0, 'No cuda support'"

Numerical Simulation

# train a model on data with style-change probability 0.75 and with statistical and 
# causal dependencies, and save it to the directory "models/mlp_example"
$ python main_mlp.py --model-id "mlp_example"     \
                     --style-change-prob 0.75     \
                     --statistical-dependence     \
                     --content-dependent-style

# evaluate the trained model
$ python main_mlp.py --model-id "mlp_example" --load-args --evaluate

Image/Text Experiment

# download and extract the dataset
$ wget https://zenodo.org/record/7678231/files/m3di.tar.gz
$ tar -xzf m3di.tar.gz

# train a model with encoding size 4 and save it to the directory "models/imgtxt_example"
$ python main_imgtxt.py --datapath "m3di" --model-id "imgtxt_example" --encoding-size 4

# evaluate the trained model
$ python main_imgtxt.py --datapath "m3di" --model-id "imgtxt_example" --load-args --evaluate

BibTeX

If you find this project useful, please cite our paper:

@article{daunhawer2023multimodal,
  author = {
    Daunhawer, Imant and
    Bizeul, Alice and
    Palumbo, Emanuele and
    Marx, Alexander and
    Vogt, Julia E.
  },
  title = {Identifiability Results for Multimodal Contrastive Learning},
  booktitle = {International Conference on Learning Representations},
  year = {2023}
}

Acknowledgements