We run this code on an Ubuntu 20.04.5 server equipped with an Intel(R) Xeon(R) Gold 6230R CPU and 8 Nvidia(R) RTX 3090 GPUs. And the packages can be found in the file requirements.txt.

You can download the datasets from here (Please contract [email protected] and complete the form to acquire Ground Truth images).

• prepare the image data in datasets/dunhuang/{}/, e.g., datasets/dunhuang/train_cond/1.png

  • train_cond: damaged images (training set)

  • train_ref: ground truth (training set)

  • test_cond: damaged images (testing set)

  • test_ref: ground truth (testing set)

This repository can also reproduce the results via the bash script provided:

• run train script ./train.sh

• run test script ./test.sh

You can download the checkpoints from here and put it in a specific folder, then set "resume_state": "checkpoint_path" in config/test.json.