geekyutao / sigma

SIGMA: Solve Image Inpainting with Guidance from Masked Autoencoders

• Deploy Environment

  conda env create -f env.yaml
  pip install -r requirements.txt

  We use PyTorch=1.7.0

• Download the trained models

  1. Link: https://drive.google.com/drive/folders/11hzes3BsYjwu39KDtIARPhMMNErgCEHJ?usp=sharing
  2. (optional, if you can't use the first link) link：https://rec.ustc.edu.cn/share/f8bd5d10-b0ca-11ec-aeed-4522c4269f96
  3. download the models and place them in current project's folder ./. It should be like ./released_model/places/inp.pth

• Prepare Val/Test Data

  1. Place the val datasets into folder structure like: path_to_val/FFHQ/val/7_kinds_mask_types/000000.png
  2. Similarily, place the test datasets into folder structure like: path_to_test/FFHQ/test/7_kinds_mask_types/000000.png
  3. The path_to_val and path_to_test is important and may be the same path.

• Execute Inference

  The running scripts for the inference process are in test_script.sh file