Your Pre-trained Diffusion Model Secretly Knows Restoration

1. 1Recognize that pre-trained diffusion models have inherent restoration capabilities
2. 2Unlock the restoration behavior in these models without requiring fine-tuning or Control-Net style modules
3. 3Apply the pre-trained diffusion models to All-in-One Restoration (AiOR) tasks for improved perceptual quality and generalization
4. 4Evaluate the performance of the pre-trained diffusion models on various restoration tasks to understand their limitations and potential applications

AI researchers and engineers working on image and signal restoration tasks can benefit from this discovery, as it simplifies the restoration process and improves efficiency

💡 Pre-trained diffusion models have secret restoration capabilities that can be unlocked for improved performance

💡 Pre-trained diffusion models can do restoration without fine-tuning!