SpMT

PyTorch Code for arXiv "Semi-parametric Makeup Transfer via Semantic-aware Correspondence"

Requirements

• Ubuntu 18.04
• Anaconda (Python, Numpy, PIL, etc.)
• PyTorch 1.7.1
• torchvision 0.8.2

Prepare data

For training phase:

opt.dataroot=MT-Dataset
├── images
│   ├── makeup
│   └── non-makeup
├── parsing
│   ├── makeup
│   └── non-makeup
├── makeup.txt
├── non-makeup.txt

For testing phase:

• Use images of MT dataset:

opt.dataroot
├── images
│   ├── makeup
│   └── non-makeup
├── parsing
│   ├── makeup
│   └── non-makeup
├── makeup_test.txt
├── non-makeup_test.txt

• Use arbitrary images:

opt.dataroot
├── images
│   ├── makeup
│   └── non-makeup
├── makeup_test.txt
├── non-makeup_test.txt

Facial masks of an arbitrary image will be obtained from the face parsing model (we borrow the model from https://github.com/zllrunning/face-parsing.PyTorch)

Train:

python train.py --phase train

Test:

1. Check the file 'options/demo_options.py', change the corresponding cofigs if needed

2. Create folder '/checkpoints/makeup_transfer/'

3. Download the pre-trained model from Google Drive and put it into '/checkpoints/makeup_transfer/'

Use images of MT dataset:

python demo.py --demo_mode normal 

Notice:

• Available demo mode: 'normal', 'interpolate', 'removal', 'multiple_refs', 'partly'

• For part-specific makeup transfer(opt.demo_mode='partial'), make sure there are at least 3 reference images.

• For interpolation between multiple references, make sure there are at least 4 reference images.

Use arbitrary images:

python demo_general.py  --beyond_mt

Results

Shade-controllable

• Interpolation from light to heavy

• Interpolation between multiple references

Part-specific

Transfer different parts from different references

Makeup Removal

Comparison with Prior Arts

• Normal Images

• Wild Images

Acknowledgments

This code borrows some function from SPADE and SCGAN