Dual-app Bridge

Deterministic Image-to-Image Translation via Denoising Brownian Bridge Models with Dual Approximators

Bohan Xiao, Peiyong Wang, Qisheng He, Ming Dong**

Note

The code and documentation here are primarily based on https://github.com/xuekt98/BBDM. We sincerely thank the authors for their contribution.

Requirements

conda env create -f environment.yml
conda activate dual-bridge

Updates

The conditional setup (conditional on Y) is highly recommended, as in our experience, it leads to faster model convergence.

Data preparation

Paired translation task

For datasets that have paired image data, the path should be formatted as:

your_dataset_path/train/A  # training reference
your_dataset_path/train/B  # training ground truth

After that, the dataset configuration should be specified in config file as:

dataset_name: 'your_dataset_name'
dataset_type: 'custom_aligned'
dataset_config:
  dataset_path: 'your_dataset_path'

Train and Test

Specify your configuration file

Modify the configuration file based on our templates in configs/Template-*.yaml

Visualizing

You can refer to visualize.ipynb for visualizing the model's sampling process. We will release additional checkpoints in future updates. However, please note that we did not conduct an in-depth exploration of the model’s hyperparameters, as this was not our primary focus.