Reproduce “EBDM: Exemplar-guided Image Translation with Brownian-bridge Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Eungbean Lee, Somi Jeong, Kwanghoon Sohn
EBDM introduces an exemplar-guided image translation framework using Brownian-bridge diffusion models, which generates photo-realistic images by efficiently transferring both global and fine-grained style from an exemplar while preserving structural control inputs. The framework demonstrates superior detail transfer and improved computational efficiency over prior methods and other diffusion-based approaches.
Key insights
- Re-conceptualizing exemplar-guided image translation as a Brownian bridge diffusion process inherently simplifies conditioning by using the structure control as a fixed initial point.
- A two-pronged approach, combining a DINOv2-based Global Encoder for coarse style and an Exemplar Network with an Exemplar Attention Module for fine-grained texture, is effective for comprehensive style transfer.
- Avoiding computationally expensive dense correspondence matching, a common bottleneck in prior methods, enhances versatility and reduces processing overhead.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.