Portrait segmentation

If you have questions, please create an issue. Polishing this repo is on my TODO list, but it's not a priority.

References

Papers

• Facetracking: Face Alignment through Subspace Constrained Mean-Shifts (Saragih et al.) (2009)
• FCNs: Fully Convolutional Networks for Semantic Segmentation (Shelhamer et al.) (2014, updated 2016)
• Portrait segmentation: Automatic Portrait Segmentation for Image Stylization (Shen et al.) (2016)

Code

• FaceTracker (Kyle McDonald)
• PyTorch train-valid-test split DataLoaders (kevinzakka)
• PyTorch FCN implementations (Kentaro Wada)

Data

• Flickr portrait-mask dataset from Shen et al. 2016 (mirror provided by timbe16)

Pretrained models

• Weights for PyTorch FCN8s implementation (provided by Kentaro Wada) (includes VGG16 ImageNet weights implicitly)

Tutorials

• PyTorch - Using Datasets and DataLoaders
• PyTorch - Defining Datasets
• Fast.ai - Deep learning course for coders

Licenses

In general: Anything I've created here is under MIT. Anything others have created is under various licenses.