BridgeTwist: Twisting Schrödinger Bridges for Training-Free Conditional Sampling
What it is
BridgeTwist is a training-free conditional sampler for pretrained Schrödinger bridge (SB) generative models. It uses a tempered twisted sequential Monte Carlo scheme with a closed-form plug-in surrogate and a bridge-shaped inference-time diffusion schedule to sample from the posterior distribution without retraining the base model.
The problem it solves
Pretrained unconditional generative models (specifically Schrödinger bridges) cannot easily be adapted to new conditional tasks without expensive retraining or specialized inverse problem solvers.
What you could use it for
- Class-conditional image generation using a pretrained unconditional SB model
- Text-prompted image generation
- Image inpainting
- Image deblurring
- Super-resolution
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.