This repository is an offical implementation of the paper "Implicit Diffusion Models for Continuous Super-Resolution" from CVPR 2023.

This repository is still under development.

The pre-trained model for 8X face continuous SR has been updated in link.

The codes are based on python3.7+, CUDA version 11.0+. The specific configuration steps are as follows:

1. Create conda environment

   conda create -n idm python=3.7.10
   conda activate idm

2. Install pytorch

   conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=11.3

3. Installation profile

   pip install -r requirements.txt
   python setup.py develop

Firstly, download the datasets used.

• FFHQ | CelebaHQ
• DIV2K ｜ Flick2K

Then, resize to get LR_IMGS and HR_IMGS.

python data/prepare_data.py  --path [dataset root]  --out [output root] --size 16,128 -l

The pre-trained checkpoints can be found at the following: link.

Run the following command for the training and validation:

sh run.sh

Add the command "-use_ddim" to implement DDIM sampling.

This code is mainly built on SR3, stylegan2-ada-pytorch, and LIIF.