Cross-Domain Image-to-Image Translation with VQGAN and BBDM

This project applies the architecture of the Brownian Bridge Diffusion Model (BBDM) in cross-domain image-to-image translation on the CelebAMask-HQ Dataset. The diffusion process is performed in the latent space of the VQGAN. This work extends the research presented in this paper by adding various commands, training on the CelebAMask-HQ Dataset, and incorporating a visualization demo.

Usage

Training

To train the model, use the following command:

python main.py --config configs/CelebAMaskHQ-f16.yaml --train --sample_at_start --save_top --gpu_ids 0 --resume_model CelebAMaskHQ-f16.pth

Sampling on Test Dataset

To sample on the test dataset, use:

python main.py --config configs/CelebAMaskHQ-f16.yaml --sample_to_eval --gpu_ids 0 --resume_model CelebAMaskHQ-f16.pth

Testing on a Single Image

To test the model on a single image, use:

python main.py --config configs/CelebAMaskHQ-f16.yaml --test --gpu_ids 0 --resume_model CelebAMaskHQ-f16.pth -i "path/to/input/image" -o "output/path"

Loading the User Interface

To load the user interface, use:

python main.py --config configs/CelebAMaskHQ-f16.yaml --ui --gpu_ids 0 --resume_model CelebAMaskHQ-f16.pth

References

• Cross-Domain Image-to-Image Translation with VQGAN and BBDM

Acknowledgements

We extend our gratitude to the authors of the original paper for their foundational work.