Intro
We propose a series of powerful image processing architectures that handles a variety of low-level vision tasks, including denoising, deraining, Super-Resolution, diffusion generation, etc. The models have SOTA performance on these tasks.
News
7/2/2024: We released a powerful All-in-One image restoration model
Instruct-IPT!
6/19/2024: Our work "
Image Processing GNN" got the "Best Student Paper Runner-Up" award at CVPR2024!
6/2/2024: We released the code and weights of
U-DiTs!
Projects
Pre-Trained Image Processing Transformer (IPT)
CVPR 2021
"The first Transformer for low-level vision"
Image Processing GNN: Breaking Rigidity in Super-Resolution (IPG)
CVPR 2024
"A flexible graph solution for SR"
Best Student Paper Runner-Up (Top 10/2720)
Instruct-IPT: All-in-One Image Processing Transformer via Weight Modulation
ArXiv
"A powerful All-in-One solution on various low-level tasks"
U-DiTs: Downsample Tokens in U-Shaped Diffusion Transformers
ArXiv
"1/6 of DiT-XL FLOPs, but better performance"
