Concept
iREPA
iREPA is a technique for representation alignment intended to improve the training efficiency of diffusion models and accelerate convergence.
Timeline (last 90 days)
Oct 14
For training acceleration, MiniDog introduces iREPA and the EUPE perceptual encoder.
Unverified50%
Oct 14
iREPA is a class of representation alignment techniques that align the intermediate-layer features of a diffusion model with the features of a pretrained visual encoder (such as DINOv2) to accelerate convergence.
Unverified50%
All Facts (2)
Unverified
For training acceleration, MiniDog introduces iREPA and the EUPE perceptual encoder.
50%UnverifiediREPA is a class of representation alignment techniques that align the intermediate-layer features of a diffusion model with the features of a pretrained visual encoder (such as DINOv2) to accelerate convergence.
50%