Makeup Transformation Network Results
Input Images
Input NonMakeup 1
Input NonMakeup 2
Input NonMakeup 3
Input Makeup 1
Input Makeup 2
Input Makeup 2
Results:
Applied Makeup 1
Applied Makeup 2
Applied Makeup 3
Removed Makeup 1
Removed Makeup 2
Removed Makeup 3
Implementation of a basic CycleGAN network trained on existing makeup datasets, however, the majority of images in
these datasets are of the full-frontal face. Considered the baseline for the comparison.
View Training Progress Images
| Training Details | ||
|---|---|---|
| Model | CycleGAN | |
| Epochs | 200 | |
| Image Dimension | 256 x 256px | |
| Generator Architecture | 9 Block Resnet | |
| Discriminator Architecture | PatchGan | |
| # Makeup Images | 504 | |
| # Nonmakeup Images | 283 | |
Applied Makeup 1
Applied Makeup 2
Applied Makeup 3
Removed Makeup 1
Removed Makeup 2
Removed Makeup 3
Implementation of a basic CycleGAN network trained with existing frontal-face makeup datasets and the new collected
dataset including faces with a range of pose.
View Training Progress Images
| Training Details | ||
|---|---|---|
| Model | CycleGAN | |
| Epochs | 200 | |
| Image Dimension | 256 x 256px | |
| Generator Architecture | 9 Block Resnet | |
| Discriminator Architecture | PatchGan | |
| # Makeup Images (Frontal) | 504 | |
| # Nonmakeup Images(Frontal) | 283 | |
| # Makeup Images (Pose) | 252 | |
| # Nonmakeup Images (Pose) | 237 | |
Applied Makeup 1
Applied Makeup 2
Applied Makeup 3
Removed Makeup 1
Removed Makeup 2
Removed Makeup 3
Implementation of a basic CycleGAN network trained with only the new collected
dataset which includes faces with and without makeup in a range of pose.
View Training Progress Images
| Training Details | ||
|---|---|---|
| Model | CycleGAN | |
| Epochs | 200 | |
| Image Dimension | 256 x 256px | |
| Generator Architecture | 9 Block Resnet | |
| Discriminator Architecture | PatchGan | |
| # Makeup Images (Pose) | 504 | |
| # Nonmakeup Images (Pose) | 237 | |
Applied Makeup 1
Applied Makeup 2
Applied Makeup 3
Removed Makeup 1
Removed Makeup 2
Removed Makeup 3
Implementation of a basic CycleGAN network trained with only the new collected
dataset which includes faces with and without makeup in a range of pose. A 6-block ResNet architecture is used
for the generator.
A ResNet architecture has the following features:
View Training Progress Images
- Skip Connections
| Training Details | ||
|---|---|---|
| Model | CycleGAN | |
| Epochs | 200 | |
| Image Dimension | 256 x 256px | |
| Generator Architecture | 6 Block Resnet | |
| Discriminator Architecture | PatchGan | |
| # Makeup Images (Pose) | 252 | |
| # Nonmakeup Images (Pose) | 237 | |
Applied Makeup 1
Applied Makeup 2
Applied Makeup 3
Removed Makeup 1
Removed Makeup 2
Removed Makeup 3
Implementation of a basic CycleGAN network trained with only the new collected
dataset which includes faces with and without makeup in a range of pose. The Unet-256 architecture is used.
A ResNet architecture has the following features:
View Training Progress Images
- Skip Connections
| Training Details | ||
|---|---|---|
| Model | CycleGAN | |
| Epochs | 200 | |
| Image Dimension | 256 x 256px | |
| Generator Architecture | Unet-256 | |
| Discriminator Architecture | PatchGan | |
| # Makeup Images (Pose) | 252 | |
| # Nonmakeup Images (Pose) | 237 | |
Applied Makeup 1
Applied Makeup 2
Applied Makeup 3
Removed Makeup 1
Removed Makeup 2
Removed Makeup 3
A network trained with the InstaGAN model. The model is built on top of CycleGAN but includes the use of segmentation
masks to mark relevant instances to transform. An additional loss function encourages to transfigure only the instances.
Due to this, it might be helpful to highlight relevant facial features to assist the network in identifying the facial structure
and pose.
| Training Details | ||
|---|---|---|
| Model | InstaGAN | |
| Epochs | 200 | |
| Image Dimension | 200 x 200px | |
| Generator Architecture | 9 Block Resnet | |
| Discriminator Architecture | PatchGan | |
| # Makeup Images (Frontal) | 504 | |
| # Nonmakeup Images (Frontal) | 283 | |
| # Makeup Images (Pose) | 252 | |
| # Nonmakeup Images (Pose) | 237 | |