Scalable Unbalanced Optimal Transport with Generative Adversarial Networks

This is the accompanying code for the paper, "Scalable Unbalanced Optimal Transport with Generative Adversarial Networks" (arXiv).

Dependencies (Python 3)

• PyTorch
• NumPy
• Visdom

This code was developed on an NVIDIA GTX 1080TI GPU.

Usage

To learn a transport map from source to target, run:

python main.py -sd results > results/log.txt

To obtain the transport results, run:

python eval.py -sd results 

The results will saved in results. By default, this code trains the model on the Zebrafish data shown in Figure 6. Our pretrained transport map can be downloaded from here: (Google Drive).