Memory-efficient filter-guided diffusion with domain transform filtering

Gustavo Lopes Tamiosso, Caetano Müller, Lucas Spagnolo Bombana, Manuel M. Oliveira

Abstract: Highlights•A memory-efficient approach for structure-conditioned diffusion models.•A normalization strategy that improves structural consistency and prompt alignment.•A strategy for handling image sizes FGD cannot within the same GPU memory budget.•A faster inference scheme than FGD with gains scaling with image resolution.•A memory-saving strategy that trades space for inference time.

External IDs:doi:10.1016/j.cag.2025.104389