PaperSnB
Split and Bridge
A multimodal diffusion generation method that requires no joint training. During sampling, different modalities share an initial diffusion trajectory, and after a split point a lightweight bridging module maintains cross-modal consistency. This allows pretrained unimodal diffusion models to be reused to achieve multimodal generation.
Timeline (last 90 days)
Nov 10
The paper "Split and Bridge: Multimodal Generation via Diffusion Bridging" proposes the SnB method, which bridges pretrained unimodal diffusion models at the sampling stage rather than jointly training a unified model.
Unverified50%