BayesianFlow
Pixel-wise uncertainty estimation in Flow Matching generative models via Last Layer Laplace Approximation
BayesianFlow extends BayesDiff from diffusion models to Flow Matching, achieving 5× faster generation than DDIM at comparable quality.
A Last Layer Laplace Approximation is integrated into the U-Net to produce pixel-wise uncertainty estimates during generation, yielding interpretable confidence maps alongside generated images. Experiments on MNIST and Fashion-MNIST assess both the quality of uncertainty estimates and the computational efficiency gap between diffusion and flow matching.
🔗 GitHub