The repo provides matting module consisting of SOTA models, open-source code for training and evaluation. Besides, there are well-trained out-of-the-box matting models for human, thus you can just use these human matting models in applications without training. If want to do image matting for other things or stuffs, you need to collect images, label images and train model. I will also convey your demand to the developers.

code: https://github.com/PaddlePaddle/PaddleSeg/tree/release/2.6/Matting