A Keras implementation of the Portrait Net architecture. Original paper available at: https://www.yongliangyang.net/docs/mobilePotrait_c&g19.pdf
Python >= 3.6Tensorflow 2.xImgaugOpenCVJupyter Notebook
It is expected that the dataset is contained in one or more folders. Each folder must contain the input image with suffix _image and all the segmentation masks with suffix _mask_<nr of class> (starts at 0). An example of folder struture can be:
- Dataset 1
-- img1_image.jpg
-- img1_mask_0.jpg
-- img1_mask_2.jpg
This shows an example where the img1 has 2 segmentation masks with ids 0 and 2. The segmentation for the id class 1, as it is not provided, will be considere as a zero matrix.
Important: it is not expected the contour images, since the data generator script grab them using the find contours algorithm from OpenCV in the segmentation masks.
You can start using this project with the provided jupyter notebook containing all aditional informations.
Contributions are allways welcome to improve this repository :D