Shypd.aiThe largest open index of AI toolsU-ViT is an Open Source & Models tool that provides a PyTorch implementation of a Vision Transformer (ViT) backbone for diffusion models. It treats all inputs as tokens and employs long skip connections for efficient image generation.
Trending
About
U-ViT is an official PyTorch implementation of the paper "All are Worth Words: A ViT Backbone for Diffusion Models" (CVPR 2023). This open-source codebase offers a simple and general ViT-based architecture for image generation with diffusion models, characterized by treating all inputs—including time, condition, and noisy image patches—as tokens. It also features long skip connections between shallow and deep layers, which significantly enhance performance and convergence speed. U-ViT supports unconditional, class-conditional, and text-to-image generation tasks, achieving competitive results compared to CNN-based U-Nets. The codebase includes optimized attention computation, pretrained models, and efficient training and evaluation scripts for various diffusion model types, leveraging techniques like mixed precision training, xformers, and gradient checkpointing for memory and speed optimization.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
