This is the official PyTorch codes for the paper

Time-Aware One Step Diffusion Network for Real-World Image Super-Resolution
Tianyi Zhang¹, Zhengpeng Duan¹, Peng-Tao Jiang², Bo Li Fu², MingMing Cheng¹, Chunle Guo^1,3,†, Chongyi Li^1,3
1 VCIP, CS, Nankai University, ² vivo Mobile Communication Co. Ltd. , ³ NKIARI, Shenzhen Futian
^†Corresponding author.

⭐ If TADSR is helpful to your images or projects, please help star this repo. Thank you! 👈

• 2025.08.25 Create this repo.

• Release training and inference code
• Release Checkpoints

1. Clone repo

git clone https://github.com/zty557/TADSR.git
cd TADSR

2. Install packages

conda create -n tadsr python==3.10 -y
conda activate tadsr
pip install -r requirements.txt

Step 1: Download Checkpoints

Download the [TADSR] checkpoints and place them in the directories preset/weights.

Step 2: Prepare testing data

Place low-quality images in preset/datasets/test_datasets/. You can download RealSR, DrealSR and RealLR200 from [SeeSR], Thanks for their awesome works.

Step 3: Running testing command

bash scripts/test_tadsr.sh

Replace the [image_path] and [output_dir] with their respective paths before running the command.

Step 4: Check the results

The processed results will be saved in the [output_dir] directory.

Step 1: Prepare the training data

• Download the training datasets LSDIR.
• Following [SeeSR], you can generate the LR-HR pairs for training using.
• Using bash_data/get_tag.sh to get the paths of each HR-LR pair and their corresponding prompts, and you will receive a dataset_list.txt file in the following format.

LSDIR/HR_image/0000001.png LSDIR/LR_image/0000001.png "tag prompt of 0000001.png"
LSDIR/HR_image/0000002.png LSDIR/LR_image/0000002.png "tag prompt of 0000002.png"
LSDIR/HR_image/0000003.png LSDIR/LR_image/0000003.png "tag prompt of 0000003.png"
...

Step 2: Start train

Use the following command to start the training process:

bash scripts/train_tadsr.sh

Replace the [txt_path] with the path to the dataset_list.txt file generated by your dataset.

This project is licensed under the Pi-Lab License 1.0 - see the LICENSE file for details.

If you find our repo useful for your research, please consider citing our paper:

@misc{zhang2025timeawarestepdiffusionnetwork,
    title={Time-Aware One Step Diffusion Network for Real-World Image Super-Resolution}, 
    author={Tainyi Zhang and Zheng-Peng Duan and Peng-Tao Jiang and Bo Li and Ming-Ming Cheng and Chun-Le Guo and Chongyi Li},
    year={2025},
    eprint={2508.16557},
    archivePrefix={arXiv},
    primaryClass={eess.IV},
    url={https://arxiv.org/abs/2508.16557}, 
}

For technical questions, please contact zty557@gmail.com