Implement some light weight hair segmentation network with keras which can be used on mobile devices easily.
- [DeeplabV3plus]:MobileNetV2 as the encoder
- PrismaNet: network architecture as described in the Prisma-AI blog
- FastDeepMatting
- PrismaNet + FastDeepMatting: base PrismaNet architecture plus the feathering block in fast deep matting
- port to iOS using CoreML
- port to iOS using ncnn
- tutorial about how to serve the model using tensorflow serving and use the results for hair color prediction